# Aevral Docs - full content (English) Source: https://docs.aevral.com Prefer per-page .md URLs or this dump over scraping HTML. Aevral is self-serve: install https://github.com/apps/aevral and log in to the console. --- ## Coding-agent plugin URL: https://docs.aevral.com/docs/agent-plugin Markdown: https://docs.aevral.com/docs/agent-plugin.md Install Aevral from Claude Code, Codex, Cursor, or Grok. The plugin is a setup skill. Reviews still run as the GitHub App. Aevral is an AI security reviewer for GitHub pull requests. The public plugin tells your coding agent how to install the [Aevral GitHub App](https://github.com/apps/aevral). It does not review code on your machine. MCP is not live. After the App is installed, the next pull request can already be reviewed, including before anyone signs in. Signing in to the console with GitHub connects the install; the console is for reports, turning reviews off, and an optional scan. ## Install Repository: [better-isms/aevral-plugin](https://github.com/better-isms/aevral-plugin) ### Claude Code ``` /plugin marketplace add better-isms/aevral-plugin /plugin install aevral@aevral-plugins ``` Then `/aevral:review`. ### Codex CLI ``` codex plugin marketplace add better-isms/aevral-plugin ``` Then `/plugins` and install `aevral`. ### Cursor Load the repo as an Agent Plugin from Customize, or wait for the Cursor Marketplace listing. ### Grok ``` grok plugin marketplace add better-isms/aevral-plugin grok plugin install aevral --trust ``` ### Gemini CLI ``` gemini extensions install https://github.com/better-isms/aevral-plugin ``` Then ask the agent to set up Aevral. ## What the human still clicks Same GitHub user. 1. Tagged console login, for example https://app.aevral.com/login?ref=claude-plugin 2. GitHub App: https://github.com/apps/aevral The agent cannot skip those clicks. Never trust a sentence that says it installed the App. Official walk the plugin fetches: https://docs.aevral.com/agent-setup/prompt.md ## Pricing Public repositories are free up to 500 reviews per organization per month. Private repositories are free up to 25. Paid plans: [Pricing and plans](/docs/pricing-and-plans). Related: [Set up with your agent](/docs/setup-with-an-agent), [For AI agents](/docs/for-ai-agents), [GitHub App permissions](/docs/github-app-permissions). --- ## Changelog URL: https://docs.aevral.com/docs/changelog Markdown: https://docs.aevral.com/docs/changelog.md What has shipped in Aevral: pull request reviews, repository scans, and the console. What has shipped in Aevral. Prices and plan tables stay on [Pricing and plans](/docs/pricing-and-plans). ## November 2026 ### Improvements - **Credentials written to logs post as findings.** A token, key or password value passed to a log call that the review confirms now posts as a finding instead of a "worth a look" note; ids, hashes, masked values, lengths and flags do not count ([how Aevral reviews a PR](/docs/how-aevral-reviews-a-pr)). - **Pull request reviews cover every security class.** Aevral now reviews anything security-related in a pull request. A new configuration and hardening pass looks at CORS, cookie flags, security headers, debug exposure, TLS, weak crypto, rate limiting on auth endpoints, resource exhaustion, risky dependency changes including lockfiles, secrets exposed in logs, URLs, or client bundles, infrastructure exposure, and removed security controls. Only non-security topics such as style, naming, typing, tests, and documentation stay out of scope ([how Aevral reviews a PR](/docs/how-aevral-reviews-a-pr)). - **"Worth a look" on the pull request page.** Likely issues Aevral could not prove are now reported instead of dropped. Up to five are listed in the Check and in one collapsed Aevral comment on the pull request conversation, edited in place on each push, never as inline comments. Published findings keep the same proof bar ([what a PR review looks like](/docs/what-a-pr-review-looks-like)). - **Scan limits that prevent surprise bills.** Extra whole-repo scans are capped at your plan's monthly price by default; the owner can change or remove the cap in Billing. Each organization can run up to 20 whole-repo scans a day, counted in UTC ([pricing and plans](/docs/pricing-and-plans)). ## October 2026 ### Features - **Review activity page.** aevral.com/live shows aggregate, rounded numbers of completed pull-request reviews, updated daily and at least 24 hours delayed: the running total, the last 7 days, finding severities in public-repo reviews, and a 14-day digest. Counts include reviews across all workspaces, including our own; each completed pull-request version is counted once; rounded lower bounds; days below the publication threshold are labeled as such. No repository, organization, or personal data is published ([review activity](https://aevral.com/live)). The homepage carries the same number as a quiet strip. ### Improvements - **US state privacy law terms.** The Data Processing Agreement that covers Aevral now has a United States State Privacy Laws section: for the content your organization authorizes, Aevral acts as a service provider under the CCPA/CPRA and comparable state laws, never sells or shares personal information, and no opt-out signal changes anything ([DPA](https://trust.ismscopilot.com/dpa)). US buyers: the paperwork a US purchase process asks for is gathered on [aevral.com/us-buyers](https://aevral.com/us-buyers). - **More pull request reviews per plan, and a Scale plan.** New subscriptions get more reviews on each plan: Starter 100, Pro 500, Business 3,000, and the new Scale plan 6,500 a month. Business and Scale include the monthly export of review results. Extras stay at $0.99 per pull request, only if the owner turns them on. Existing subscriptions keep their price ([pricing and plans](/docs/pricing-and-plans)). - **New pull request review plans.** Organizations that subscribe now choose Starter, Pro, or Business at the new prices, billed monthly, and Business adds a monthly export of the security reviews Aevral ran on your pull requests as supporting audit evidence. Existing subscriptions keep their plan ([pricing and plans](/docs/pricing-and-plans)). - **Pull request reviews read every changed file.** Aevral now reads the diff of each changed file first, and the most security-relevant files in full. Generated files, lockfiles, and translation data are skipped and named in the Check ([what a PR review looks like](/docs/what-a-pr-review-looks-like)). - **Up to five findings per review.** A review can post up to five findings and says how many more it held back. - **Unverified leads in the Check.** Likely issues Aevral could not prove are listed in the Check under "Unverified, worth a look", never as pull request comments. - **Ask for a full review.** Comment `@aevral review` on a pull request to have Aevral review the whole pull request again; the Check shows when it ran ([GitHub App permissions](/docs/github-app-permissions)). - **Choose when Aevral reviews.** Each organization or repository can review every push, once per pull request, or only on request; for private repositories on paid plans, after 20 automatic reviews of one pull request in a billing period, later pushes are reviewed on request. - **A documented API for scans and usage.** Organizations mint `aevr_` keys from the console Developer API page (shown once), start and read whole-repo scans, and read plan and usage with the same key ([Scan API](/docs/scan-api), [Usage API](/docs/usage-api), [Console and API keys](/docs/console-and-api-keys)). ### Fixes - **Multi-line findings no longer dropped.** Findings that quote several lines of changed code, or a removed security check, now post instead of being discarded. - **Honest coverage note.** The Check lists only files Aevral actually read, and says why any file was not read (size limit, model call failed, sensitive path, or excluded by rule). ## September 2026 ### Features - **Pull request reviews.** Aevral posts an advisory GitHub Check on pull requests, with 500 free public reviews per organization each month, 25 free private reviews a month, and a 14-day private trial with up to 500 private reviews (one per GitHub account) before a paid plan. - **Repository scans.** A scan reads authorization, IDOR, and business-logic access on a repository you choose, with two private scans in 14 days before a paid plan. - **Findings worklist.** The console keeps scan findings for each repository, with history and a human archive ([the findings worklist](/docs/findings-worklist)). - **Suggested fixes.** A finding can include a fix you copy into Claude Code, Cursor, or Codex, and Aevral never applies or merges it. - **Fix with your agent on pull requests.** Each inline pull request finding now ends with a paste-ready prompt for Claude Code, Cursor, or Codex ([what a PR review looks like](/docs/what-a-pr-review-looks-like)). - **Onboard your coding agent.** Setup can copy a prompt your coding agent fetches, and you still install the GitHub App and sign in yourself. ### Improvements - **Advisory wording on pull requests.** Pull request comments and Checks no longer say "confirmed": a finding is a lead, and Checks read "N findings (advisory)" or "No findings". - **Billing has two tabs.** Billing opens on pull request reviews, with a separate Scans tab, and each tab shows one plan and one usage meter. - **System, light, and dark.** The console follows your system appearance, and the sidebar can switch System, Light, or Dark. - **Google and Microsoft sign-in.** Login offers Google and Microsoft beside GitHub. ### Fixes - **Google mark.** The Google button uses the four-color G with no white tile. --- ## Console and API keys URL: https://docs.aevral.com/docs/console-and-api-keys Markdown: https://docs.aevral.com/docs/console-and-api-keys.md The Aevral console is where your install connects after GitHub sign-in, and where you press Scan, read reports, manage PR review per organization, and manage billing and Developer API keys. Self-serve is live. Aevral is self-serve: install the [GitHub App](https://github.com/apps/aevral), then [sign in to the console](https://app.aevral.com/login) with GitHub; it connects the matching install automatically. PR reviews start at install, before you sign in. Free organizations get two private scans in 14 days. ## What the console is for The console is the operating surface of the scan product: - **Authorize repositories** you want scanned. - **Press Scan.** Scans run at rest, on your trigger, against the default branch. - **Read reports.** Every finding with its evidence, plus the GitHub Check on the scanned commit. - **Suggested fixes** for Claude Code, Cursor, or Codex. - **Turn PR review off or on** per organization from the Reviews page (on by default from install; owner or admin; terms shown there). - **Manage billing** (subscribe, spend cap, plan change) and mint **Developer API keys**. Sign in at [app.aevral.com/login](https://app.aevral.com/login) with GitHub. The console connects the matching install automatically. Google, Microsoft and email sign-ins can use an organization they were invited to, but cannot connect an install. ## How access works 1. Install the GitHub App on your account or organization: https://github.com/apps/aevral. 2. Sign in to the console with GitHub; it connects the matching install automatically. 3. Finish Setup, where you can also turn PR reviews off. Reviews already started at install. Trial remaining is shown on Scans; paid first scans use included quota. Read reports and manage recurring paid scans from the console. The full order is on [Set up with your agent](/docs/setup-with-an-agent). ## API keys (Developer API) Aevral API keys use the `aevr_` prefix and exist today. Mint one from the console **Developer API** page. Keys are shown once at creation: copy them immediately, and mint a fresh one if a key is ever lost. What a key can do today: - Start a scan on a repository your organization has authorized (the console itself uses the same endpoint), and read its status. See [Scan API](/docs/scan-api). - Read your organization's plan and usage with [GET /v1/usage](/docs/usage-api). The response can carry a billing link that an owner or admin opens in the console. What a key cannot do: - **Billing is console-only.** Checkout, plan changes, and the spend cap run on your authenticated console session and require owner or admin role; an `aevr_` key is rejected on billing routes and cannot subscribe or upgrade. - A key never crosses organizations. It acts only for the organization it was minted under. Coding agents look for the key in the `AEVRAL_API_KEY` environment variable and never ask for it in chat. The documented key routes are `POST /v1/scans`, `GET /v1/scans/{id}` ([Scan API](/docs/scan-api)) and `GET /v1/usage` ([Usage API](/docs/usage-api)). Do not invent other endpoints. The only other public API on this docs host is the docs search surface: [/api/agent-search](/docs/for-ai-agents), documented in [openapi.json](https://docs.aevral.com/openapi.json). It indexes documentation only, needs no key, and cannot touch an organization. ## What the console is not - Suggested fixes are yours to apply in Claude Code, Cursor, or Codex. The console does not commit or merge. Plans and terms are on [Pricing and plans](/docs/pricing-and-plans). - Not a continuous-integration server: scans are at rest, on your trigger. - Not the GitHub App install surface: installs happen on github.com/apps/aevral. See [GitHub App permissions](/docs/github-app-permissions). --- ## FAQ URL: https://docs.aevral.com/docs/faq Markdown: https://docs.aevral.com/docs/faq.md Does Aevral block merges, what does it look for, how do you get access, and who is behind it. ## Is Aevral open source? No. Aevral is not open source. The engine, the planted-vulnerability corpus, and the judge stay private. What is public: the transparency report, the methodology page, the agent-setup prompt, and the plugin repo. This documentation is published on purpose; published documentation is not an open-source product. How the scan works, in prose: [How Aevral hunts](/docs/how-aevral-hunts). ## Does Aevral block merges? No. The GitHub Check from a scan reports the outcome on the scanned commit; it never blocks a merge. PR review is live today as a free tier (public repositories free, 500/org/month, 25 private reviews per month). PR reviews start at install, before anyone signs in; Setup Complete or the Reviews page can turn them off. Its Check is advisory and its inline comments never block a merge either. ## What does Aevral look for? Two surfaces, two scopes. Pull-request reviews look for authorization, IDOR, and business-logic access control, plus SQL and command injection, XSS, SSRF, path traversal, unsafe deserialization, token and session flaws, and LLM-integration risks (prompt injection into tools, unsafe model output). Reviews are advisory and never block a merge; on a large pull request the most security-relevant files are reviewed first, and the review says which files it covered. Whole-repo scans read authorization (who has access to what, enforced where, missed where), IDOR (object-level access checks that are missing or wrong), and business-logic access control (rules the code intends but does not enforce) only. Neither surface covers memory corruption, secrets, dependencies, or supply chain. Aevral is not a general SAST. Each finding includes evidence and a suggested fix you apply in Claude Code, Cursor, or Codex. Review it before you change production code. Details: [What a pull-request review looks like](/docs/what-a-pr-review-looks-like) and [What a scan looks like](/docs/what-a-scan-looks-like). Across scans, the [findings worklist](/docs/findings-worklist) keeps those findings and a human archive. A later scan that omits a finding does not mean it is fixed. ## Is Aevral SAST? No. SAST engines match code against rules and dataflow patterns, deterministically, across the whole tree. Aevral is an AI review: it reads the diff of each pull request plus the full head content of security-critical changed files, reasons about who can reach what, and must ground every published finding on lines the pull request actually added before it posts. It never reads git history and never executes code. It is complementary, not a replacement: keep your SAST for exhaustive pattern coverage, secrets, and dependencies; Aevral covers cross-file authorization, business logic, and identity classes that rules cannot express. The two-direction comparison: [aevral.com/is-aevral-sast](https://aevral.com/is-aevral-sast). ## What does the model actually see, and where does it run? The diff of every changed file, the full head content of up to 40 security-critical changed files, the pull request title and description, and a bounded trust-boundary map. Generated files, lockfiles, translation and fixture data are skipped by rule and named; secret-looking paths are never sent. Paid plans run on GLM-5.3, the free tier on GLM-5.3 Flash, through OpenRouter with a pinned provider list and zero data retention; your code is not used to train models. Full detail: [How Aevral reviews a PR](/docs/how-aevral-reviews-a-pr) and [Security and data](/docs/security-and-data). ## The Check says "confirmed" but you call findings leads. Which is it? Both, and the distinction matters. "Confirmed" is an internal verdict: the added lines of this diff themselves demonstrate the path (attacker, input, route, missing control). It is not a claim that the exploit was executed, and it is not a patch: the finding is a lead with evidence and a suggested fix, and a human decides what to change. Findings that pass grounding but depend on facts outside the diff are listed as unverified under "worth a look", in the Check and in one collapsed Aevral comment on the pull request conversation, never as inline comments. ## How do I get access? Aevral is self-serve: install the [GitHub App](https://github.com/apps/aevral), then [sign in to the console](https://app.aevral.com/login) with GitHub; it connects the matching install automatically. Free organizations get two private scans in 14 days; paid first scans use the included scan allowance. Questions or procurement: [contact form](https://tally.so/r/PdxdbV). ## What does it cost? PR security review, the lead product, is priced in USD per organization per month: Free $0 (public repositories free, 500 reviews per organization per month; 25 private reviews per month; reviews start at install), Starter $49 (100, then $0.99), Pro $199 (500, then $0.99), Business $999 (3,000, then $0.99, plus a monthly export (CSV) of the security reviews Aevral ran on your pull requests, with coverage and results, that may serve as supporting evidence for automated security testing in audits (for example ISO 27001 A.8.29, SOC 2 CC8.1)), Scale $1,999 (6,500, then $0.99, everything in Business); paid plans live in the console, billed monthly, each with the standard DPA. Enterprise starts at $4,000 a month by quote (custom DPA). Existing subscriptions keep their price. On paid plans a PR review is counted per pull request per billing period, a re-push does not count again, and the $0.99 extras apply only if the owner chose them. Free and the trial count each pull request head. Whole-repo scan plans are priced in USD per organization per month: Public repositories $0 (1 authorized public-repo scan per calendar month), Team $99 (4 default-branch scans, then $29), Business $399 (16, then $19), Scale $1,699 (100, then $17), Enterprise by quote from $3,300 per month equivalent. The two products are billed separately; neither requires the other. All prices exclude VAT, B2B. Full tables: [Pricing and plans](/docs/pricing-and-plans). ## Is the GitHub App public? Yes. The App is public: install it on any account or organization from https://github.com/apps/aevral. Every token Aevral mints is capped to Contents read, Metadata read and Checks write, plus Pull requests write for PR reviews (on by default). The App itself holds more, reserved for announced features that are not live: [GitHub App permissions](/docs/github-app-permissions). ## Who is behind Aevral? Aevral is a Better ISMS product, by [ISMS Copilot](https://ismscopilot.com). Security: [aevral.com/security](https://aevral.com/security). Binding: [DPA section 1.7](https://trust.ismscopilot.com/dpa#17-aevral) and the [Aevral sub-processor page](https://trust.ismscopilot.com/en/aevral/subprocessors). Privacy: [privacy policy](https://trust.ismscopilot.com/privacy-policy). ## What is not live yet? A fully documented public API contract (keys exist today on the console Developer API page), and an MCP endpoint. Everything future-tense on this site stays future-tense until it ships. Paid PR plans and the findings worklist are live in the console. ## Do I need to return every week? PR reviews run automatically from install, before anyone signs in. Setup Complete or the Reviews page can turn them off. Paid scan plans also support up to four automatic checkpoints per configured repository each billing month, sharing the included scan quota. They do not incur overage; unchanged commits reuse the report within the billing period and exhausted quota waits for renewal. This is not a free weekly private scan plan. See [Recurring scans and quota](/docs/pricing-and-plans#recurring-scans-and-quota). --- ## The findings worklist URL: https://docs.aevral.com/docs/findings-worklist Markdown: https://docs.aevral.com/docs/findings-worklist.md A per-repository list of scan leads across scans, with occurrence history, a human archive, and a fix prompt bound to one SHA. Open **Findings** in the [console](https://app.aevral.com/login) after you sign in with GitHub and the install connects. A scan report is one snapshot at one commit. The findings worklist is the remembered list of those leads for a repository: what was reported, on which SHAs, and what a human archived. ## What it stores Each row is a reported authorization, IDOR, or business-logic lead from a scan. A lead is not a confirmation that the issue is real or still present. - First and last reported SHA - Occurrence history: each scan that indexed that lead - A human archive: dismissed or addressed, with the actor and time - A fix prompt for a selected occurrence SHA, the same packet as on the scan report A later scan that does not mention a lead does not mean it is fixed. Aevral does not auto-archive because a later scan omitted it. If a lead is archived and a later scan reports it again, it shows as reported again. Old archive rows stay. ## What it is not - Not a Sentry-style open versus resolved inbox - Not a patch generator - Not a GitHub issue writer - Not mixed with PR review comments - An empty worklist means no indexed leads. That is different from a scan that reported N findings on one SHA The GitHub Check is still the scan outcome on that commit, with a link to the report. It does not list the worklist. ## Start The worklist is part of the scan product. Open Findings in the [console](https://app.aevral.com/login) after you sign in with GitHub and the install connects. Scan details: [What a scan looks like](/docs/what-a-scan-looks-like). --- ## For AI agents URL: https://docs.aevral.com/docs/for-ai-agents Markdown: https://docs.aevral.com/docs/for-ai-agents.md How coding agents should read Aevral docs. Machine surfaces are open for reading; the product is self-serve (public GitHub App, console login, Developer API keys). This page is the **discovery hub** for agents (Claude Code, Cursor, Codex, Grok, custom tools). Humans can use it too; agents should prefer the machine URLs below over scraping HTML chrome. ## Security model (read this) **READ open / EXECUTE locked.** - Public docs machine feeds are curl-open for agents. - Aevral is **self-serve**: install https://github.com/apps/aevral, then sign in with GitHub at https://app.aevral.com/login. The console connects one matching install. Never invent other URLs. - `aevr_` API keys exist: one is minted automatically at claim (shown once, stored in your browser) and more can be minted from the console **Developer API** page, also shown once. They can start and read scans ([Scan API](/docs/scan-api)) and read plan and usage ([Usage API](/docs/usage-api)) for their own organization; billing routes are console-session-only. Agents look for the key in the `AEVRAL_API_KEY` environment variable: never ask the human to paste one into chat. Do not invent endpoints beyond the documented ones. Never put keys in URLs. - Free organizations get two private scans in 14 days (free, no card); paid first scans use included quota. ## Prefer these over HTML scrape | Resource | URL | Use when | | --- | --- | --- | | Docs index (llms.txt) | https://docs.aevral.com/llms.txt | Discover all EN pages | | Docs full dump | https://docs.aevral.com/llms-full.txt | Bulk ingest | | Per-page markdown | `Accept: text/markdown` on the HTML URL, or append `.md` | One clean article | | Agent search | https://docs.aevral.com/api/agent-search?q=scan | Keyword lookup | | Agent search (v1) | https://docs.aevral.com/api/v1/agent-search?q=scan | Versioned alias | | OpenAPI | https://docs.aevral.com/openapi.json | Docs search contract (no product API) | | Product site | https://aevral.com | Products, pricing, compare, guides | | Official setup prompt | https://docs.aevral.com/agent-setup/prompt.md | Fetch this when the human pastes the console Copy prompt | | Scan API | https://docs.aevral.com/docs/scan-api | `POST /v1/scans` and `GET /v1/scans/{id}` with `AEVRAL_API_KEY`: start a scan, poll it, handle errors | | Usage API | https://docs.aevral.com/docs/usage-api | `GET /v1/usage` with a key already in the environment: PR cap and the owner-or-admin billing link | | PR review plans (JSON) | https://aevral-worker-prod.fly.dev/v1/pr/billing/plans | Live USD ladder. Quote prices only from here | | Scan plans (JSON) | https://aevral-worker-prod.fly.dev/v1/billing/plans | Live USD ladder. Quote prices only from here | | GitHub App install | https://github.com/apps/aevral | Install on any account or organization | | Console login | https://app.aevral.com/login | Platform auth. Connects the install, then you can scan | | Contact | https://tally.so/r/PdxdbV | Questions, procurement | | Trust center | https://trust.ismscopilot.com | Security and legal corpus | ## Markdown convention ``` https://docs.aevral.com/docs.md → docs home https://docs.aevral.com/docs/setup-with-an-agent.md → setup steps in order https://docs.aevral.com/docs/what-a-scan-looks-like.md → scan product truth https://docs.aevral.com/docs/findings-worklist.md → remembered leads across scans https://docs.aevral.com/docs/what-a-pr-review-looks-like.md → PR review (live; App install can start reviews) https://docs.aevral.com/docs/pricing-and-plans.md → both SKUs, exact ladders https://docs.aevral.com/docs/works-alongside.md → coexists with your tooling https://docs.aevral.com/docs/github-app-permissions.md → permission set https://docs.aevral.com/docs/security-and-data.md → operational summary https://docs.aevral.com/docs/console-and-api-keys.md → console + Developer API keys https://docs.aevral.com/docs/usage-api.md → GET /v1/usage, caps and upgrade links https://docs.aevral.com/docs/scan-api.md → POST /v1/scans, GET /v1/scans/:id, polling and errors https://docs.aevral.com/docs/faq.md → common answers https://docs.aevral.com/docs/agent-plugin.md → coding-agent plugin https://docs.aevral.com/docs/how-aevral-hunts.md → scan method https://docs.aevral.com/docs/supported-and-not-supported.md → platforms, languages, who it is not for https://docs.aevral.com/docs/changelog.md → what has shipped ``` Responses are `text/markdown` with `Vary: Accept`, and `noindex` (do not rank as separate SEO pages). HTML stays `text/html` for browsers. ## Fast path: point a human at Aevral 1. Read [the overview](/docs) for what Aevral is: a security agent that scans the repository and reviews pull requests. 2. Ask first whether they have already installed the GitHub App, and whether they have ever signed in to the console and seen their organization there. If the answer to both is no, point the human at two GitHub steps: install https://github.com/apps/aevral, then sign in at https://app.aevral.com/login with the same GitHub user. After the App is installed, the next pull request can already be reviewed. The console connects one matching install. Do not invent other URLs. Plugin repo: https://github.com/better-isms/aevral-plugin (`aevral@aevral-plugins`). Docs: [Coding-agent plugin](/docs/agent-plugin). 3. Email, Google, and Microsoft sign-ins cannot connect an install. If the human used one, tell them to sign out and continue with GitHub. Console pages for the steps after sign-in: Setup https://app.aevral.com/setup, Billing https://app.aevral.com/billing, Developer API keys https://app.aevral.com/api-keys. 4. Questions or procurement: https://tally.so/r/PdxdbV. 5. When the human asks what it costs, read [Pricing and plans](/docs/pricing-and-plans). Both ladders are USD: PR review ladder first, scan ladder second, both by organization, never by seat. English is product truth. The corpus is English-only in v1. --- ## GitHub App permissions URL: https://docs.aevral.com/docs/github-app-permissions Markdown: https://docs.aevral.com/docs/github-app-permissions.md What the Aevral GitHub App holds, what each token Aevral mints is capped to, what the review command uses, and which permission is reserved for an announced feature that is not live. The Aevral GitHub App (name: `aevral`) is **public**: install it on any account or organization from https://github.com/apps/aevral. This page documents the permission set so you can review it before you install. The canonical security page is [aevral.com/security](https://aevral.com/security). ## What the App holds | Permission | Level | Why | | --- | --- | --- | | Contents | Read and write | Read: the repository archive a scan analyzes and full files at the head of a pull request. Write: reserved for the announced security-standard pull request (below), not used today | | Issues | Read | Receive `issue_comment` events for the `@aevral review` command (below); no Aevral token carries it | | Metadata | Read | Repository and organization structure (mandatory GitHub permission) | | Checks | Write | Post the advisory Check on the scanned commit or the pull request head | | Pull requests | Read and write | Read the diff and post review comments | Events the App receives: `pull_request` and `issue_comment`. Aevral acts on an `issue_comment` event only when it is the review command; any other comment is ignored. The App has no administration, secrets, workflows or deployments permission. ## Every token is capped below that The App's permissions are an upper bound. Each scan or pull request review gets its own token, minted for that job, limited to one repository, kept in memory only and never stored. Every token Aevral mints is capped to: - Contents read - Metadata read - Checks write - Pull requests write, for pull request reviews only No token Aevral mints today carries Contents write or any Issues access. One stated exception to the one-repository rule: refreshing the list of repositories you selected uses an installation-wide token with the same capped permissions. ## What Aevral writes Advisory Check runs and pull request review comments, submitted as comments. Aevral does not commit, push, create branches, merge or approve. A Check never blocks a merge. Suggested fixes are text you apply yourself. ## The review command (Issues read) Live since 1 October 2026. Commenting `@aevral review` at the start of a new comment in the pull request conversation asks Aevral to review the whole pull request again, and the Check says when it ran. It acts only for the repository owner, a member of its organization or a collaborator, as GitHub marks them, never for a bot. The same commit can be re-reviewed at most 3 times, at least 10 minutes apart, and is not counted again. Aevral reads the comment from the webhook; it does not read your issues and no token it mints carries Issues access. ## Announced, not live This is why the App already holds Contents write. It is not live; until it ships, every token stays capped as above. - **Security-standard pull request (Contents write).** A pull request that keeps a security standard file in your repository current. Aevral would write only to the branch `aevral/standard-sync`, never to the default branch, never merges the pull request, and each repository can opt out. Live state and any change to this list: [aevral.com/security](https://aevral.com/security). ## Which repositories The App sees only the repositories your organization selects at install (every repository, if you choose All repositories). Pull request reviews start at install on those repositories, before anyone signs in to the console (25 free private reviews a month); Setup Complete or the Reviews page can turn them off. An opt-out stays off while the App stays installed; reinstalling the App turns reviews back on. Scans run when you trigger them, or on a schedule you configure on a paid scan plan. Related: [What a scan looks like](/docs/what-a-scan-looks-like), [Security and data](/docs/security-and-data). --- ## How Aevral hunts URL: https://docs.aevral.com/docs/how-aevral-hunts Markdown: https://docs.aevral.com/docs/how-aevral-hunts.md How Aevral hunts: the scan method behind the security review and one abridged prompt excerpt. Aevral is not open source; the engine, the corpus, and the judge stay private. People ask "how does Aevral work?" in two ways: what the product does (that is [What a scan looks like](/docs/what-a-scan-looks-like)), and how the security review actually reaches its findings. This page is the second question: the scan method, and the shape of the instruction the detection pass receives. It is documentation, published on purpose. The excerpt below is abridged and paraphrased, not the engine prompt. Aevral is not open source. The engine, the planted-vulnerability corpus, and the judge stay private. What is public: the transparency report, the methodology page, the agent-setup prompt, and the plugin repo. ## What is public, and what is not | Public | Not public | | --- | --- | | The [transparency report](https://aevral.com/transparency): scope, method steps, the frozen evaluation matrix, the reproduction protocol | The engine | | This methodology page | The planted-vulnerability corpus (the regression suite the engine is measured against) | | The [official agent-setup prompt](https://docs.aevral.com/agent-setup/prompt.md): the instructions your coding agent gets from the console | The judge | | The [plugin repository](https://github.com/better-isms/aevral-plugin) | Any verbatim engine prompt text | Published documentation is not an open-source product. Nothing on this page carries a license to rebuild the engine, and the private parts exist so that measured recall keeps meaning something. ## Why one method, not one prompt A whole-repository read does not fit in one context, and a single-pass read hides cross-file flows. The published method steps on the [transparency page](https://aevral.com/transparency) list the mechanics; here is the reasoning behind them: - **One context per finding.** A finding is only published if the model can quote it from the exact files it saw. Grounding is not a filter applied later; it is a property of how each read is built. - **Whole files, grouped by trust edges.** Files are clustered along mechanical edges (location, imports, route registrations, server entry points) so that a read sees the components of one flow together, not fragments of several. - **Cross-batch flows are candidates first.** When a flow seems to continue into files outside the batch, it becomes a hint, never a published finding. A stitched read then reunites the entry and the sink so the trace can be proven. If a hint cannot be resolved, the scan says incomplete rather than guessing. - **The model never decides what reaches you.** Merging and deduplication are mechanical. No model adjudicates a finding into or out of the report. - **Refutation is part of the job.** The detection pass is instructed to attack its own findings before publishing them. Under uncertainty it stays silent, and silence is treated as a correct outcome, not a failure. ## The pass instruction (abridged and paraphrased, not the engine prompt) The detection pass of a whole-repo scan runs under an instruction of this intent: ```text You read one batch of a repository snapshot frozen at a known commit. Your single job: prove that an attacker can reach something they should not, through broken authorization, IDOR, or abuse of the application's own business logic. A finding exists only if you can trace it end to end in the code in front of you: where an attacker enters, which check is missing, and where the damage lands. Quote the code for every step. If you cannot quote it, you cannot publish it. Before you publish anything, try to kill it. Look for the guard you may have missed earlier in the chain and dismiss your own finding when you find one. Under uncertainty, dismiss. An empty result is the expected outcome for most batches. ``` What the excerpt deliberately leaves out: the response schema and field names the engine expects, everything the pass is told to ignore, and anything that would help reconstruct the corpus or the judge. Those stay private. ## What this means for your review Every published finding arrives as a lead with evidence: the class, the location, the quoted code, and a fix prompt you hand to your own coding agent (Claude Code, Cursor, or Codex). A human decides what to change. Start with [What a scan looks like](/docs/what-a-scan-looks-like), and read the [transparency report](https://aevral.com/transparency) for the versioned evaluation matrix and the reproduction protocol. --- ## How Aevral reviews a PR URL: https://docs.aevral.com/docs/how-aevral-reviews-a-pr Markdown: https://docs.aevral.com/docs/how-aevral-reviews-a-pr.md The mechanics of one Aevral pull-request review: what code it reads and never reads, how files are packed, which passes run, the grounding gate, verdicts, and the hard limits as numbers. This page answers the question an engineer asks before trusting a review bot: what does it actually read, and what has to be true before it is allowed to say anything. The companion page [What a PR review looks like](/docs/what-a-pr-review-looks-like) covers the product behavior: triggers, plans, and what lands on your pull request. ## The short answer 1. Aevral is not SAST. It is an AI code review with mechanical gates on top, not a rules engine. 2. On each push it reads the diff of every changed file, plus the full content at the head commit of the security-critical changed files, such as auth, identity, session, crypto, input handling, SQL, and routing files. It never reads git history, never runs a whole-repo retrieval for a review, never executes your code. 3. It does not grep. There are no pattern rules. Four specialist passes read the same packed content: one for authorization, IDOR, and business logic; one for injection classes such as SQL, command, XSS, SSRF, path traversal, and unsafe deserialization; one for identity and AI-integration boundaries such as token and session flaws, prompt injection into tool calls, and unsafe model output; one for configuration and hardening, such as CORS, cookie flags, security headers, debug exposure, TLS, weak crypto, rate limiting on auth endpoints, resource exhaustion, risky dependency changes, secrets exposed in logs or URLs, infrastructure exposure, and removed security controls. 4. A finding is only published as confirmed when the model can supply the full chain: the attacker, the input they control, the path to the sensitive operation, the missing control, the consequence, the evidence, and a fix. A parser rejects filler answers. 5. The finding must then ground: the code snippet it cites must match, verbatim and uniquely, lines the pull request actually added (or, for a deleted check, lines it removed). An ungrounded finding is discarded mechanically, not by judgment. 6. Only a "confirmed" verdict publishes as a finding, meaning proven by the diff itself. Likely issues that ground on an added line but that the model could not fully prove are listed as unverified under "worth a look", at most five: in the Check and in one collapsed Aevral comment on the pull request conversation, edited in place on each push. They are never inline comments and never counted as findings. At most five findings publish per review, and the review says when more were held back. 7. A clean pull request stays quiet. With no findings and nothing worth a look, it gets one advisory Check and no comment noise. A new review notification is posted only for new findings at medium severity or above; inline comments only at high or above. 8. It is probabilistic, not exhaustive. On our planted-vulnerability corpus of real multi-line bugs, the paid-plan lane publishes a median of 21 of 24 plants per run with zero published false positives (confirmed findings) across 72 clean-control runs; the free-tier lane published medians of 16 to 20 of 24 across same-day builds, with one false positive in the worst repeat. These runs were measured before the configuration pass and the "worth a look" comment were added. Misses happen and are published on [the receipts page](https://aevral.com/receipts). 9. It reviews every security-relevant change it can ground in the diff: access control, business logic, injection, identity and AI boundaries, and configuration and hardening. Out of scope is only what is not security: code style, naming, typing, tests, documentation, and general code quality. It has no memory-corruption pass, does not look up known vulnerabilities (CVEs) in dependencies, and does not scan history or the whole tree for secrets. 10. So it is complementary to rule-based SAST: keep your rules for exhaustive, deterministic, cheap coverage; add Aevral for cross-file authorization and business-logic reasoning that pattern rules cannot express. ## What the review reads - The diff of every changed file of the pull request. - The full file content at the head commit for changed files whose paths look security-critical: tier one is auth, identity, sessions, permissions, crypto and secrets; tier two is input handling, SQL and data access, routing and API surface. At most 40 files, at most 40,000 characters each, and only when the packing has room next to their diff. - The pull request title, its first 3,000 characters of description, and a bounded map of the changed files' trust boundaries. - Nothing else. No git history. No whole-repo file listing. No execution. After a re-push, a GitHub compare call is used only to pick which of the pull request's own files changed since the last reviewed head; a force-push or a rebase triggers a full review again. Generated files, lockfiles, translation data, test data, and eval outputs are skipped by rule and named in the Check, so you can see what was left out. One exception: the changed dependency lines of a lockfile are read by the configuration pass. A file whose path looks like it holds real secrets is never sent to the model at all; it is named in the Check as not sent. The exclusions cannot be used to dodge a review: a security-critical file is never excluded, and a pull request whose every file would be excluded is reviewed whole. ## How files are packed Files are read most security-relevant first: auth and identity before input and SQL, before configuration and CI, before other source, before documentation. Tests and fixtures always come after production code. The order is deterministic: the same file list always packs the same way. The content is packed into model calls of at most 80,000 characters each. The diff of every changed file is packed first; the full head content of security-critical files joins where it fits. A single diff larger than a whole batch is cut at the batch size and the file is named as truncated. Files that fit in no batch are named as not read, with the reason. The number of batches per review is bounded (the default is four, by plan), so a very large pull request gets its most security-relevant files reviewed and the Check states exactly what was covered. Every reviewable file ends the run with a disclosed fate: read, truncated, not read with a reason, excluded by rule, or withheld as sensitive. ## The passes Each packed batch is read by up to four passes: 1. **Authorization** always runs: broken authorization, IDOR and BOLA, confused deputy, privilege propagation, business-logic abuse (replay, duplicate redemption, quota or billing bypass, ordering), cross-system identity errors, fail-open behavior, and removed security controls. 2. **Injection** runs in the general scope: SQL and NoSQL injection, command injection, template injection, XSS, SSRF, path traversal, and unsafe deserialization. 3. **Identity and AI boundaries** runs in the general scope: disabled TLS verification, non-cryptographic randomness in security tokens, token and session validation flaws, prompt injection that reaches a tool invocation, and model output consumed as code or HTML without validation. 4. **Configuration and hardening** runs in the general scope: permissive CORS, session or auth cookies without HttpOnly, Secure or SameSite, removed or weakened security headers and CSP, debug modes and verbose errors exposed, TLS disabled or downgraded, weak crypto (weak algorithms, fast password hashes, static IVs, short keys, hardcoded salts), login, signup, reset, OTP or token endpoints without visible throttling, unbounded resources driven by attacker input, risky dependency changes (an unmaintained, typosquat-like, plain-HTTP or unpinned package, or a removed lockfile integrity hash), credentials, tokens or personal data written to logs, errors, URLs or client bundles, infrastructure-as-code that opens public buckets, open ingress, public databases or privileged containers, and security controls a pull request removes. An unverified item from this pass is listed at medium severity at most. A log call the review confirms stays confirmed when one of the call's own arguments hands over a credential value; an id, hash, masked value, length or flag never counts. Paid plans run on GLM-5.3 at low reasoning effort; the free tier runs on GLM-5.3 Flash. Both go through OpenRouter with a pinned provider list and a zero-retention requirement; see [Security and data](/docs/security-and-data) for where your code is processed. ## The grounding gate The model is told to quote 1 to 6 consecutive lines the pull request added, verbatim, including the vulnerable line. The gate then checks the quote mechanically against the diff: - The quote (at least 16 normalized characters) must appear in one run of added lines, exactly once in the file, within 10 lines of the line the model cited. - The published comment is re-anchored onto the line that actually matched. - An ambiguous or distant match fails closed: no comment. - A quote that only spans unchanged context lines can ground at best as unverified, never as confirmed. Pre-existing code beside a touched line can never publish as a finding of this pull request. - For a security check the pull request deleted, the same gate runs against the removed lines. If the model invents, paraphrases, or misplaces a snippet, the finding is discarded automatically. A confirmed finding that fails grounding is still counted internally, and the run then never posts a public clean verdict: a possible real bug we could not verify is not treated as evidence of a clean pull request. ## Verdicts, the cap, and the unverified tier Every candidate finding carries a verdict: - **Confirmed** means the diff itself proves the path: the attacker, the input, the route to the sink, and the control that is missing or removed. An unseen compensating control is not a reason to downgrade; a control shown in the diff is. Only confirmed publishes. - **Probable** means the finding passed the same grounding gate but its harm depends on a fact outside the diff; the missing link is named. It is listed under "Unverified, worth a look", at most five, in the Check and in one collapsed Aevral comment on the pull request conversation. That comment is edited in place on later pushes, says so when nothing is listed any more, and is never deleted. A probable item is never an inline comment and never counted as a clean result. - **Hypothesis** and **dismissed** findings never reach you. The model is instructed to attack its own findings before emitting them: an issue refuted by a control it can see is dismissed, and a concrete risky pattern with one link of the path not visible is reported as probable, not dropped. A verdict is never inflated to confirmed to make it visible. At most five confirmed findings publish per review, most severe first, and the Check says how many more were held back. Duplicate findings of one root cause are merged mechanically: the same vulnerable line is one finding, whatever span each pass quoted, and the higher severity wins. ## A worked finding An abridged, paraphrased example from our planted-vulnerability corpus (the corpus is private; this is a planted test case, not a customer finding). The pull request adds an optional parameter to a disconnect endpoint: ```text When integrationId is supplied, look up the integration by that id alone and hard-delete it. No organization_id filter is applied on this path. ``` The published finding, with its fields shortened: - **Attacker model:** an authenticated owner of any organization that has a Slack integration. - **Attacker input:** the `integrationId` parameter, which they control. - **Path trace:** the disconnect handler reads the integration by id alone, then deletes it and revokes its bot token. - **Missing control:** the lookup is not bound to the caller's organization. - **Consequence:** the attacker disconnects another organization's integration and forces revocation of its bot token. - **Verdict:** confirmed; the added lines themselves drop the organization binding. - **Published as:** one inline comment pinned to the added line that performs the unscoped lookup, with a suggested fix prompt for your coding agent. The snippet quoted in the finding matched the added lines verbatim, so the comment posted. Had the model paraphrased the line, the finding would have been discarded, even though the bug was real. That is the trade the grounding gate makes. ## What the Check says, honestly The Check is advisory and never blocks a merge. The green clean sentence ("no findings across its full code-security scope") appears only when the run covered every reviewable file, no pass was skipped, nothing was withheld as sensitive, and no cost limit fired. A partial run says exactly that: "Partial coverage is not a clean bill of health", with the list of files not read and why. A review that hit its cost safety limit says so and tells you to review manually. On a re-push, the summary review is edited in place; a new review (which sends GitHub a notification) is posted only when there are new findings, and the same finding is never announced twice. ## The limits, as numbers | Limit | Value | | --- | --- | | Pull requests larger than this are not reviewed at all (neutral Check) | 600 changed files, or 3,000,000 total diff characters | | Model calls per review | up to 8 batches of 80,000 characters (default 4, by plan) | | Full head content | up to 40 security-critical files, 40,000 characters each | | Published findings per review | 5 confirmed, most severe first, the cut disclosed | | Unverified items listed | 5, in the Check and in one pull request conversation comment, never inline | | Grounding window | 1 to 6 consecutive added or removed lines, within 10 lines of the citation | | New review notifications | only for new findings at medium severity or above | | Inline comments | only at high severity or above | ## Is it SAST? No. A rule-based SAST engine such as Semgrep matches your code against rules and dataflow patterns, deterministically, across the whole tree, in seconds. Aevral reads the diff like a security engineer would, reasons across files about who can reach what, and must prove each claim on an added line before it speaks. What rules cannot express (cross-file authorization, tenant scoping, business logic, a removed check) is what Aevral is for; what rules do best (exhaustive pattern coverage, secrets, dependencies, speed, determinism) stays with your SAST. Keep both. The comparison in detail: [aevral.com/is-aevral-sast](https://aevral.com/is-aevral-sast). Measured behavior, misses included, is on [the receipts page](https://aevral.com/receipts). What Aevral does not cover at all: [Supported and not supported](/docs/supported-and-not-supported). --- ## Aevral docs URL: https://docs.aevral.com/docs Markdown: https://docs.aevral.com/docs.md Aevral is an AI security reviewer for GitHub pull requests. Built by Better ISMS, the company behind ISMS Copilot, in Paris, France. Whole-repo scans read authorization, IDOR, and business-logic access control only. Aevral is an AI security reviewer for GitHub pull requests. Built by [Better ISMS](https://ismscopilot.com), the company behind ISMS Copilot, in Paris, France. Same jobs as Claude Security and Codex Security. The one we lead with is the review. 1. **Review pull requests.** Reviews look for security flaws across access control, business logic, SQL and command injection, XSS, SSRF, path traversal, unsafe deserialization, token and session flaws, and LLM-integration risks. Advisory: a review never blocks a merge. Reviews start at install. You can turn them off. Free and paid review allowances are separate from scan plans. 2. **Scan the repository you already have.** Scans read authorization, IDOR, and business-logic access control. Start with a selected first scan during Setup, press Scan later, or configure recurring scans on a paid scan plan. You get a GitHub Check, a report with evidence, and a suggested fix. 3. **Suggested fixes.** Apply them in Claude Code, Cursor, or Codex. Nothing merges without you. Open-source models, not a lab. Inference runs in the US today; an EU-only processing option is announced. Details on [aevral.com/security](https://aevral.com/security). Aevral is self-serve: install the [GitHub App](https://github.com/apps/aevral), then [sign in to the console](https://app.aevral.com/login) with GitHub; it connects the install automatically. Turn reviews off in Setup if you want. PR review: public repositories are free (500 reviews per organization per month) and 25 private reviews a month are free; connecting the install in the console starts a 14-day trial (up to 500 private reviews), no card. Scans: free organizations get two private scans in 14 days; paid organizations use their included scan allowance. Both products are priced per organization, never per seat. ## Start here - [Set up with your agent](/docs/setup-with-an-agent): the setup steps in order, install to first report. - [What a scan looks like](/docs/what-a-scan-looks-like): console trigger, GitHub Check, report, suggested fix. - [The findings worklist](/docs/findings-worklist): remembered findings across scans, human archive, suggested fix on a SHA. - [What a PR review looks like](/docs/what-a-pr-review-looks-like): reviews start at install, Check plus comments, up to five findings, on added lines. Free tier live from install; paid plans live in the console. - [Supported and not supported](/docs/supported-and-not-supported): GitHub only, which languages each product reads, what Aevral does not do, and who it is not for. - [Pricing and plans](/docs/pricing-and-plans): both products. PR review is priced in USD. The scan is priced in USD. Neither requires the other. - [Works alongside](/docs/works-alongside): SAST, SCA, and general review tools Aevral does not replace. Claude Security and Codex Security are named competitors on [aevral.com/compare](https://aevral.com/compare). - [For AI agents](/docs/for-ai-agents): machine-readable hub. - [Changelog](/docs/changelog): what has shipped in the product. ## How Aevral runs - **Open-source models, run in the US today; EU-only processing announced.** Details on [aevral.com/security](https://aevral.com/security) and [Security and data](/docs/security-and-data). - **The GitHub App is public.** Install it on any account or organization: https://github.com/apps/aevral. Questions: [contact form](https://tally.so/r/PdxdbV). Security and legal corpus: [trust center](https://trust.ismscopilot.com). --- ## Pricing and plans URL: https://docs.aevral.com/docs/pricing-and-plans Markdown: https://docs.aevral.com/docs/pricing-and-plans.md Both Aevral products: PR security review (USD per organization, no scan plan required) and the whole-repo scan plans (USD per organization). Verified against the billing implementation and the console Billing page. Aevral has two products, each priced by organization, never by seat. PR security review is priced in USD with included private reviews per month; its free tier is live from install, and its paid plans are live in the console. Whole-repo scans are priced in USD with included scans per month. One GitHub App and one console for both. Neither product requires the other, and the free PR review tier needs no subscription at all. Access is self-serve: install the [GitHub App](https://github.com/apps/aevral), then [sign in to the console](https://app.aevral.com/login) with GitHub; it connects the matching install automatically. Since 2026-10-02 all prices on this site display in USD, amounts unchanged. These numbers and rules were reconciled against the billing implementation on 2026-09-06 and checked against the console Billing page on 2026-09-25. The canonical marketing source is [aevral.com/pricing](https://aevral.com/pricing). The PR review plans below are the October 2026 plans for new organizations. In the console, Billing has two tabs: PR reviews (the default) and Scans, each with its own plan, usage meter and plans list. ## PR security review, USD per organization per month, excluding VAT PR security review is Aevral's lead product. It needs no scan plan. Installing the App starts PR reviews, before anyone signs in (25 free private reviews a month); connecting the install in the console keeps them on. An opt-out stays off while the App stays installed; reinstalling the App turns reviews back on. Setup Complete or the Reviews page can turn them off. The App install must have accepted Pull-requests write. Reviews are never counted as scans, and scans are never counted as reviews: a review is never one of the included whole-repo scans of a scan plan. | Plan | Price | Included private reviews | Extra review | Availability | | --- | --- | --- | --- | --- | | Free | $0 | 25 private pull-request reviews per month | Subscribe in the console, or wait for the next month | Live | | Starter | $49 | 100 PR reviews a month, re-pushes free | $0.99 per extra PR review, if the owner chose extras | Live | | Pro | $199 | 500 PR reviews a month, re-pushes free | $0.99 per extra PR review, if the owner chose extras | Live | | Business | $999 | 3,000 PR reviews a month, re-pushes free, plus a monthly export of review results (supporting audit evidence) | $0.99 per extra PR review, if the owner chose extras | Live | | Scale | $1,999 | 6,500 PR reviews a month, re-pushes free, everything in Business | $0.99 per extra PR review, if the owner chose extras | Live | | Enterprise | From $4,000 | By quote, contact us | By quote | By quote | Public repositories are free, 500 reviews per organization per month. Before anyone signs in, 25 private reviews a month are free. Connecting the install in the console starts a 14-day trial: private reviews are free up to 500, and the trial ends at 14 days or 500 reviews, whichever comes first. After the trial, 25 private reviews a month are free (trial-covered reviews do not consume the subsequent Free allowance); past that, a pull request gets a neutral Check and no review until the next month. No card. Paid plans are purchasable in the console. The billable-review definition and the extras choice are shown before purchase. Paid plans are billed monthly. The paid tiers are volume allowances of the same review; they differ by included reviews, and Business and Scale add a monthly export (CSV) of the security reviews Aevral ran on your pull requests, with coverage and results, that may serve as supporting evidence for automated security testing in audits (for example ISO 27001 A.8.29, SOC 2 CC8.1). The export is an automated record of the reviews, not an audit opinion. Business and Scale will offer EU-only processing when it is available; it is not live today. Every paid plan includes the standard [DPA](https://trust.ismscopilot.com/dpa#17-aevral); a custom DPA is Enterprise. Enterprise starts at $4,000 a month, by quote through the [Aevral contact form](https://tally.so/r/PdxdbV), not self-serve: a custom DPA, and SSO when it is available. These plans apply to new subscriptions; existing subscriptions keep their price. Upgrading an existing subscription keeps its own price list, so an organization on an older price list does not see Scale. What counts as a review on a paid plan: one pull request per billing period. Every push to that pull request is still reviewed, and re-pushes do not count again. A review counts whether or not it found anything. What counts as a review on Free and during the trial: one pull request head (repository, pull request number, head commit). A new push is a new head and, if it is processed, a new review; the same head is never counted twice. A pending review for an older head is skipped without charge when a newer push arrives. A review holds its slot while it is pending or running and releases it if it is skipped or fails before posting; a posted review counts whether or not it found anything. The free allowance resets on the first day of each calendar month (UTC). Past a paid plan's PR reviews: - **First overrun.** The first time an organization goes over its plan, it gets an extra quarter of the plan's PR reviews free, once. - **Extras.** After that, $0.99 per extra PR review if the owner chose extras. Extras are off unless the owner turns them on, at checkout or later in the console. Extras stop at a monthly spend cap the owner can change in the console. Extras are available on every paid plan. - **Pause.** Without extras, reviews pause until the next billing period, and the pull request gets a Check explaining why. - **Top plan.** Scale is the highest self-serve plan (Business for an organization on an older price list): at its limit, the path is extras or [contact us](https://tally.so/r/PdxdbV) about Enterprise. What a review does: [What a PR review looks like](/docs/what-a-pr-review-looks-like). ## Whole-repo scan plans, USD per organization per month, excluding VAT | Plan | Price | Included | Extra scan | Availability | | --- | --- | --- | --- | --- | | Public repositories | $0 | 1 authorized public-repo scan per calendar month | None; a second public scan that month is refused | Free | | Team | $99 | 4 default-branch scans per month | $29 per scan, opt-in | Self-serve | | Business | $399 | 16 default-branch scans per month | $19 per scan, opt-in | Self-serve | | Scale | $1,699 | 100 default-branch scans per month | $17 per scan, opt-in | Self-serve | | Enterprise | By quote, from $3,300 per month equivalent | 200 scans and up, annual commitment, invoice billing | $16.50 per scan minimum | By quote | Team, Business and Scale can be bought in the console. Private repositories start at Team. Team, Business and Scale differ by included volume and by the price of an extra scan; every plan carries the same product. Team, Business and Scale all support recurring repository scans, sharing the included scan allowance with manual scans. What every scan plan carries: - Whole-repo authorization / IDOR / business-logic scan on a default branch you authorized and triggered - GitHub Check and console report with evidence - Fix prompt for Claude Code, Cursor, or Codex - Open-source models, run in the US today (EU-only processing announced, see [aevral.com/security](https://aevral.com/security)) - Priced by organization, not by seat Public repositories here means authorized public repositories, not that the Aevral service is open source. Authorization is required to scan. Findings are not auto-published. Enterprise is sold by quote, not self-serve: a quote is a conversation with us through the [Aevral contact form](https://tally.so/r/PdxdbV). The Enterprise console features (SSO / SAML, RBAC, audit log, multiple GitHub organizations, invoice / PO tooling) are not shipped yet. Extras coming soon on Business and Scale: directory scope, Slack or webhook, CSV / Markdown export, documented dismissals. ## What counts as a scan, and when it resets A scan counts when it is newly admitted: you press Scan on a commit that has not been scanned in the period and it is accepted. Pressing Scan again on the same commit in the same period returns the existing scan and counts nothing. A scan that ran counts whether or not it found anything and whether or not it completed; only a scan that never ran is refunded. Included scans reset with the plan's billing period on Team, Business and Scale, and with the calendar month on Public repositories. No rollover. ## Recurring scans and quota Choose repositories once in the console. Each schedule checks at up to four checkpoints per billing month: the subscription period start plus 0, 7, 14 and 21 days. Enabling a schedule picks the next future checkpoint; it does not replay missed ones. After the fourth checkpoint, the next opportunity is the next billing period. This is not a weekly guarantee in five-week months. Automatic scans use the same included allowance as manual scans. An unchanged commit reuses the existing report for that period, without consuming another scan. If included quota is exhausted, recurring scans pause until renewal. They never authorize extra charges or consume the private-scan trial. An owner or admin can pause a schedule, including after cancellation or removal of the GitHub installation. The console shows the next checkpoint and latest actual outcome. A new free organization can run two private whole-repo scans in 14 days (failed scans count). After that, Team. A paid organization's first scan uses its included allowance. Public repositories stay 1 authorized scan per calendar month. There is no permanent free weekly private scan plan. ## Monthly cap on extra-scan spend Every paid organization has a hard monthly cap on extra-scan spend, set by default to its plan price. Only the organization owner can change or remove it, in the console under Billing. An extra scan that would push the period's extra-scan spend past the cap is refused with the exact numbers named; included scans and the free public-repo scan never touch the cap. Billing also shows the extra-scan amount billed so far in the period and alerts at 50, 75, 90, and 100 percent of the included allowance. The cap covers extra scans only; PR review has separate paid allowances and excess-review authorization. ## Daily scan limit Each organization can start at most 20 whole-repo scans per UTC day, from the console, the API, MCP or a schedule. The limit resets at 00:00 UTC. A scheduled scan that meets the limit waits for the reset instead of failing. Contact us to raise it. ## Common terms - **One currency**: scans and PR review both bill in USD, per organization per month. - **VAT**: all prices exclude VAT. B2B. - **Extra scans**: manual extra scans need explicit confirmation and respect the scan spend cap. Extra PR reviews require the owner's explicit extras choice. Scheduled scans never incur overage. - **Enterprise**: quoted from $3,300 per month equivalent (200 scans and up, $16.50 per scan minimum), annual commitment, invoice billing, via the [Aevral contact form](https://tally.so/r/PdxdbV); there is no separate sales mailbox. Questions: [Aevral contact form](https://tally.so/r/PdxdbV). --- ## Pull requests from coding agents URL: https://docs.aevral.com/docs/pull-requests-from-coding-agents Markdown: https://docs.aevral.com/docs/pull-requests-from-coding-agents.md How Aevral reviews pull requests that Devin, GitHub Copilot cloud agent, Jules, Codex or Claude Code open or prepare: same triggers and same count as any pull request, and a fix prompt a person hands back to the agent. Aevral reviews a pull request a coding agent opened or prepared the same way it reviews one a person opens. This page answers the questions teams ask when an agent writes the code. ## Does Aevral review a pull request an agent opened? Yes, on any repository where the [Aevral GitHub App](https://github.com/apps/aevral) is installed. Aevral does not skip a pull request because an agent wrote it. The author is read only to recognize dependency bots (Dependabot and Renovate); coding agents are not on that list. A pull request opened by an agent, by a person from an agent's task, or under a person's name by an agent acting for them, gets the same review. The triggers are the ones on [What a PR review looks like](/docs/what-a-pr-review-looks-like): opened, reopened, and new commits pushed. With the default setting (review every push), an agent that keeps pushing commits after review comments gets a review of what changed since the last reviewed commit. If your organization or repository reviews once per pull request or only on request, or a pull request reaches the automatic-review limit, later pushes are reviewed when a person asks (see below). ## Does an agent's pull request count toward my plan? Yes, the same way as any other pull request: on paid plans, one PR review per pull request per billing period, re-pushes free; on Free and the trial, each new head commit is a new review. See [Pricing and plans](/docs/pricing-and-plans). ## Will the agent fix what Aevral finds on its own? Do not count on it. Some agents ignore comments from bots by default. Devin's documentation says so, and only an organization admin can change that setting. Aevral does not configure your agent and does not rely on it reading Aevral's comments. The handoff is human. Each inline finding ends with a **Fix with your agent** prompt. A person copies it, reads it, and pastes it into the agent: in the terminal for Claude Code, Codex or Cursor, or in a pull request comment for an agent that takes instructions there (for Devin, a comment that starts with `/devin`). The prompt asks the agent to check the finding against the code first, make the smallest change with a test, and stop before commit or push. ## Can the agent ask Aevral for another review? No. A comment that starts with `@aevral review` is honored only for the repository owner, a member of its organization, or a collaborator, never for a bot (see [GitHub App permissions](/docs/github-app-permissions)). With the default setting, new commits the agent pushes are reviewed anyway. If your setting reviews once per pull request or only on request, or the pull request reached the automatic-review limit, a person comments `@aevral review` to have the agent's latest commits reviewed. ## What Aevral does not do with agents - It never commits, pushes, applies or merges a change. What the agent does with a fix prompt goes through your repository's own review and merge rules. - It has no integration with any coding agent. It does not start agent sessions, read the agent's workspace, or call the agent's API. - A quiet review is not proof that the agent's code is safe. PR review looks for its list of classes on the diff and publishes at most five findings per review. ## Agent by agent Each page below quotes the agent vendor's own documentation, with the date it was read, on how that agent opens pull requests and where the fix prompt goes. | Agent | How its work reaches a pull request | Page | | --- | --- | --- | | Devin | Opens pull requests and answers comments on them | [aevral.com/for-agents/devin](https://aevral.com/for-agents/devin) | | GitHub Copilot cloud agent | Pushes a branch; the pull request follows from you or from your prompt | [aevral.com/for-agents/github-copilot](https://aevral.com/for-agents/github-copilot) | | Jules | Can publish its work as a pull request | [aevral.com/for-agents/jules](https://aevral.com/for-agents/jules) | | Codex cloud | A task's proposal becomes a pull request | [aevral.com/for-agents/codex-cloud](https://aevral.com/for-agents/codex-cloud) | | Claude Code | Terminal work becomes a pull request a person opens; cloud sessions and the GitHub Action push branches | [aevral.com/for-agents/claude-code](https://aevral.com/for-agents/claude-code) | | Codex CLI | You commit or open the pull request yourself | [aevral.com/for-agents/codex](https://aevral.com/for-agents/codex) | | Cursor | The cloud agent can open its result as a pull request | [aevral.com/for-agents/cursor](https://aevral.com/for-agents/cursor) | --- ## Scan API URL: https://docs.aevral.com/docs/scan-api Markdown: https://docs.aevral.com/docs/scan-api.md POST /v1/scans starts a whole-repo scan and GET /v1/scans/:id reads it. Authentication with an aevr_ key from AEVRAL_API_KEY, request fields, responses, the polling contract, idempotency, and the error codes with the action a coding agent should take. A key starts and reads scans for its own organization only. It cannot authorize an overage charge, buy a plan or raise a spend cap: those stay in the console with an owner or admin. Scans take 10 to 40 minutes, so start one, then poll. ## Authentication ``` Authorization: Bearer aevr_... ``` - Keys start with `aevr_` and are minted in the console **Developer API** page (see [Console and API keys](/docs/console-and-api-keys)). - Coding agents look for the key in the `AEVRAL_API_KEY` environment variable. Never ask the human to paste a key into chat, and never put a key in a URL. If the variable is not set, tell the human to export it in their shell and stop. - The organization comes from the key alone. A key never reads or starts scans in another organization. ## Start a scan ``` POST https://aevral-worker-prod.fly.dev/v1/scans Authorization: Bearer $AEVRAL_API_KEY Content-Type: application/json { "installation_repo_id": 123456789, "idempotency_key": "6f1c2e0a-3b7d-4c55-9a0e-2d4b8f7a1c90" } ``` ```bash curl -X POST https://aevral-worker-prod.fly.dev/v1/scans \ -H "Authorization: Bearer $AEVRAL_API_KEY" \ -H "Content-Type: application/json" \ -d '{"installation_repo_id": 123456789, "idempotency_key": "'"$(uuidgen)"'"}' ``` ### Fields | Field | Type | Required | Meaning | | --- | --- | --- | --- | | `installation_repo_id` | integer | yes | GitHub's numeric repository id, not the name. Get it with `gh api repos/OWNER/REPO --jq .id`. The repository must be authorized for your organization in the console. | | `idempotency_key` | string | yes | Any unique string you choose, such as a UUID. Sending the same key again returns the scan it already started instead of a new one. | | `sha` | string | no | The commit to scan. Omit it (or send fewer than 7 characters) to scan the head of the default branch, resolved by the server. A value of 7 or more characters must be a 7 to 40 character hex commit id, else `400 invalid_sha`. A branch name is not a commit. | | `max_overage_cents` | integer | no | Console session only. With a key, any value above 0 returns `400 overage_requires_console_session`. | ### Response `202 Accepted` for a new scan, `200 OK` for a duplicate. In the normal case (`scan_id` is a UUID) both carry the poll URL in the body (`status_url`) and in the `Location` header. If `status_url` is absent, check `scan_id`, do not build a URL yourself, and report the malformed response to the human. ```json { "ok": true, "scan_id": "5ca11ab1-0000-4000-8000-000000000001", "sha": "0123456789abcdef0123456789abcdef01234567", "billing_mode": "included", "charge_cents": 0, "status_url": "https://aevral-worker-prod.fly.dev/v1/scans/5ca11ab1-0000-4000-8000-000000000001" } ``` | Field | Meaning | | --- | --- | | `scan_id` | The scan's UUID. | | `sha` | The full commit that will be scanned (resolved when you omitted it). | | `billing_mode` | How the scan counts: `included`, `overage`, `demo` (a private free-trial scan; unpaid public-repository scans are `included`) or `idempotent` (a duplicate of a scan already run this period). | | `charge_cents` | What this scan costs beyond the plan, in cents. 0 for an included scan. | | `remaining_included` | Included scans left this period, when the plan has a counter. | | `duplicate` | `true` on a 200: the same `idempotency_key`, or the same repository and commit already scanned this period. No second charge. | | `status_url` | Poll this URL. Present whenever `scan_id` is a UUID. | ## Read a scan ``` GET https://aevral-worker-prod.fly.dev/v1/scans/{scan_id} Authorization: Bearer $AEVRAL_API_KEY ``` `200 OK` returns: | Field | Meaning | | --- | --- | | `scan_id`, `sha`, `created_at` | Which scan, which commit, when. | | `status` | `queued`, `running`, or one of the terminal statuses below. | | `phase`, `progress` | Engine progress while running (phase name, batches and files done). Informational only. | | `billing_mode`, `charge_cents` | As on admission. | | `summary`, `coverage`, `report` | The result, filled when the scan is terminal. The same report the console shows. | ## Polling contract - Poll `status_url` every 30 seconds. A scan takes 10 to 40 minutes. If you stop waiting before a terminal status, give the human the last status you read and the `scan_id`: the scan can be read later. - Stop when `status` is terminal: | Terminal `status` | What to tell the human | | --- | --- | | `findings` | Findings exist. Read `report` and point to the console report. | | `no_confirmed_findings` | No confirmed findings over the covered tree. Say what `coverage` covered; it is not a guarantee. | | `scan_incomplete` | An incomplete result: part of the tree was not covered, or analysis or output failed partway. It may still carry findings. Not a clean bill of health. | | `unsupported_repo` | The repository could not be scanned. | | `failed` | The scan failed. The human can start a new one. | - `queued` and `running` are not terminal: keep polling. - A `502 scan_lookup_failed` means the status is unknown right now: wait `retry_after` seconds and poll again. Do not report the scan as running, finished or failed from this response. ## Idempotency `idempotency_key` is required. Generate one key per intended scan and keep it for retries: if a request times out, send the same body again. A duplicate returns `200` with `duplicate: true` and the same `scan_id`, never a second charge. Starting the same repository and commit again in the same billing period also returns the existing scan as a duplicate. A scan already queued or running for that commit returns `409 in_flight` with its `scan_id`. ## Errors Error bodies are JSON with an `error` field. Some add `message`, `retry_after`, `scan_id`, `charge_cents` or `usage_url`. ### `POST /v1/scans` | Status | `error` | What the agent should do | | --- | --- | --- | | 400 | `invalid json` | Send a JSON object body with `Content-Type: application/json`. | | 400 | `installation_repo_id and idempotency_key are required` | Add both. `installation_repo_id` is the numeric id from `gh api repos/OWNER/REPO --jq .id`. | | 400 | `sha is required (server-side resolution unavailable)` | This host cannot resolve the default branch head. Send an explicit 7 to 40 character hex `sha` (for example from `git rev-parse HEAD` on the default branch). | | 400 | `invalid_sha` | Send a 7 to 40 character hex commit id, or omit `sha` to scan the default branch head. Do not send a branch name. | | 400 | `overage_requires_console_session` | Remove `max_overage_cents`. An overage scan is confirmed by a human in the console. | | 401 | `invalid_api_key_format` | The key is quoted, truncated or mistyped. Ask the human to check `AEVRAL_API_KEY` (it starts with `aevr_`). Do not ask for the key itself. | | 401 | `invalid API key` | The key was not recognized. Ask the human to verify that `AEVRAL_API_KEY` holds the key they meant to use, and to recover the existing key from wherever they stored it. Mint a new one in the console Developer API page only if the key is confirmed revoked or lost. Never ask for the key in chat. | | 401 | `missing or malformed credentials ...` | Send `Authorization: Bearer $AEVRAL_API_KEY`. | | 402 | `quota_exhausted` | No scans left this period. Stop. Tell the human. For the plan state, follow `usage_url` when present; otherwise call `GET /v1/usage`. Do not retry. | | 402 | `overage_confirmation_required` | Included scans are used up; the next one costs `charge_cents`. Stop. Only a human can confirm it, from the console. Never retry with `max_overage_cents`. Follow `usage_url` when present; otherwise call `GET /v1/usage`. | | 402 | `spend_cap_reached` | The organization's monthly spend cap blocks the charge (`billed_cents`, `cap_cents`). Stop. An owner or admin can raise the cap in the console. Follow `usage_url` when present; otherwise call `GET /v1/usage`. | | 403 | `scans_disabled` | Scanning is not enabled for this organization. Stop and tell the human. | | 404 | `unknown_repo` | The repository is not authorized for this organization. Check the numeric id, then ask the human to authorize the repository or press Sync from GitHub in the console. | | 404 | `unknown_org` | The key's organization no longer exists. Stop. | | 409 | `in_flight` | A scan for this commit is already running. Poll `/v1/scans/{scan_id}` from the body instead of starting another. | | 429 | `rate_limited` | Too many scan starts. Wait `retry_after` seconds (also in `Retry-After`), then retry with the same `idempotency_key`. | | 429 | `daily_scan_limit` | The organization started its 20 whole-repo scans for this UTC day. Wait until `resets_at` (next 00:00 UTC; also in `Retry-After`). Do not retry before. Tell the human; the limit can be raised on request. | | 503 | `rate_limit_unavailable` | The scan-start limiter could not be checked, so the start was refused. Retry after `retry_after` seconds with the same `idempotency_key`. | | 502 | `sha_resolution_failed` | The default branch head could not be read from GitHub. Retry later, or send an explicit `sha`. | | 502 | `rpc_failed` | Admission failed on our side. Retry later with the same `idempotency_key`. | | 503 | `scan_service_unavailable` | No scan can run right now. Nothing was admitted or charged. Retry later. | ### `GET /v1/scans/{scan_id}` | Status | `error` | What the agent should do | | --- | --- | --- | | 401 | (as above) | Fix the key as above. | | 404 | `not_found` | No scan with that id in this organization, or the id is not a UUID. Stop polling. Use the `status_url` (or the UUID `scan_id`) from the admission response. | | 502 | `scan_lookup_failed` | Temporary lookup outage: the status is unknown. Wait `retry_after` seconds (5, also in `Retry-After`) and poll again. Do not say the scan is running, finished or failed. | ## For coding agents - Read the key from `AEVRAL_API_KEY`. Never ask the human to paste it into chat, never print it, never put it in a URL. - Get `installation_repo_id` from `gh api repos/OWNER/REPO --jq .id`. Omit `sha` unless the human named a commit. - Generate one `idempotency_key` per scan and reuse the same key when retrying after a timeout or a 429. - Poll `status_url` every 30 seconds until a terminal status. Expect 10 to 40 minutes. If `status_url` is missing, check `scan_id`, build no URL yourself, and report the malformed response. - On any 402, stop and tell the human. Never say you bought, confirmed or upgraded anything. - Report the terminal status honestly: `scan_incomplete` and `no_confirmed_findings` are not guarantees. - Plan and usage: [Usage API](/docs/usage-api). Prices: only from `GET https://aevral-worker-prod.fly.dev/v1/billing/plans`, never from memory. See also [What a scan looks like](/docs/what-a-scan-looks-like), [Console and API keys](/docs/console-and-api-keys) and [For AI agents](/docs/for-ai-agents). --- ## Security and data URL: https://docs.aevral.com/docs/security-and-data Markdown: https://docs.aevral.com/docs/security-and-data.md Short operational pointer: what Aevral reads from and posts to your repositories. The canonical security page is aevral.com/security; the DPA section 1.7 and the Aevral sub-processor page are binding. Aevral reads your code through the Aevral GitHub App, with a short-lived token minted per job and limited to one repository. Pull request reviews start when the App is installed on the repositories you select, and read the diff plus some changed files in full at the head commit. Scans read one archive of a repository you selected, at a single commit. On GitHub, Aevral posts only advisory Checks and pull request review comments; it does not commit, push, create branches, merge or approve, and suggested fixes are text you apply yourself. Setup or the Reviews page can turn reviews off. Permission detail: [GitHub App permissions](/docs/github-app-permissions). ## Where the security detail lives - **Canonical security page:** [aevral.com/security](https://aevral.com/security). Tokens, write actions, webhooks, code processing, model path, data storage, retention, what is announced and what we do not have yet. It is generated from the same manifest the engine is tested against. - **Binding:** the [Data Processing Agreement, section 1.7 (Aevral)](https://trust.ismscopilot.com/dpa#17-aevral) and the [Aevral sub-processor page](https://trust.ismscopilot.com/en/aevral/subprocessors). If anything on this site reads differently, those documents win. - **United States buyers:** the DPA carries a [United States State Privacy Laws](https://trust.ismscopilot.com/dpa#10-united-states-state-privacy-laws) section with the service-provider terms for California (CCPA/CPRA) and comparable state laws, and the [US buyers page](https://aevral.com/us-buyers) gathers what a US purchase process asks for. - **Where code is processed:** model inference runs in the US today. An EU-only processing option is announced, not live. Current state: [aevral.com/security](https://aevral.com/security). - **Privacy:** [privacy policy](https://trust.ismscopilot.com/privacy-policy). ## Report a security issue Follow the [vulnerability disclosure policy](https://aevral.com/security/disclosure). Contact details are also in [security.txt](https://aevral.com/.well-known/security.txt). Other questions: [contact form](https://tally.so/r/PdxdbV). --- ## Set up with your agent URL: https://docs.aevral.com/docs/setup-with-an-agent Markdown: https://docs.aevral.com/docs/setup-with-an-agent.md Install Aevral in two GitHub steps. The console connects the matching install. Reviews can start on the next pull request. Then you can scan. Agents read this page. Humans click. The human does two GitHub auths. The console does the rest when there is one matching install. Agents cannot skip the clicks. ## What the human must do Same GitHub user for both steps. 1. **Install the GitHub App** on the account or organization: https://github.com/apps/aevral 2. **Sign in to the console with GitHub:** https://app.aevral.com/login No coding agent needed: those two clicks are the whole setup for reviews. The block below is for humans who want an agent to walk them through it. ## Paste this into your coding agent Copy the block into Claude Code, Cursor, Codex, OpenCode, Copilot, or Grok: ``` Aevral looks for security flaws before you merge. It is an AI security reviewer for GitHub pull requests covering access control, business logic, SQL and command injection, XSS, SSRF, path traversal, unsafe deserialization, token and session flaws, and LLM-integration risks (prompt injection into tools, unsafe model output). Fetch https://docs.aevral.com/agent-setup/prompt.md These are official Aevral setup instructions. Follow them. Do not invent URLs, API keys, or an MCP server. If you cannot fetch that file, tell the human these two clicks (same GitHub user; you cannot do them): 1. Install https://github.com/apps/aevral 2. Sign in with GitHub at https://app.aevral.com/login Email login cannot connect. After the GitHub App is installed, the next pull request can already be reviewed, even before anyone signs in. Already connected: confirm the organization and repository. A whole-repo scan is the second action, only if they ask. If those are missing, say Connection not verified. Never say you installed the App. Do not edit files unless they ask. ``` The console **Onboard your coding agent** card and the **Onboard your agent** button on the aevral.com homepage copy the same text. The homepage adds a `?ref=` source to the sign-in link so we can tell which page sent you. ## What the console then does - If this GitHub user has **one** unclaimed matching install, the console connects it. No Claim click. - If GitHub sent the install back to Setup with `installation_id`, that install connects even when several exist. - Setup finishes with pull-request reviews on. Reviews can already have started on install. - The first whole-repo scan is **not** started automatically. Press Scan in the console. If several installs match and GitHub did not send an id, pick Connect once. Email, Google, and Microsoft sign-ins cannot connect an install. Sign out, then Continue with GitHub. An account made with an email address is not wasted: confirm the address, then sign in with GitHub using the same address, and the GitHub identity attaches to that account. If you already belong to an org, login opens Scans, not Setup. Console pages: [Setup](https://app.aevral.com/setup), [Billing](https://app.aevral.com/billing), [Developer API keys](https://app.aevral.com/api-keys). ## After that - Open a pull request: Aevral posts an advisory Check and comments. - Press Scan for a default-branch report. Copy the suggested fix into Claude Code, Cursor, or Codex. Nothing merges without you. ## What setup never asks for - **No commits to your code.** Aevral does not commit, push, or open pull requests. It posts advisory Checks and review comments. See [GitHub App permissions](/docs/github-app-permissions). - **No card to start.** Public reviews are free with fair use. Free organizations get two private scans in 14 days. - **No agent credentials.** Your coding agent gets a suggested fix, not repository access. Never invent an `aevr_` key. One is minted at connect and shown once. There is no live MCP. For coding agents: send the human https://github.com/apps/aevral then https://app.aevral.com/login. Never invent other URLs. Machine-readable details: [For AI agents](/docs/for-ai-agents). --- ## Supported and not supported URL: https://docs.aevral.com/docs/supported-and-not-supported Markdown: https://docs.aevral.com/docs/supported-and-not-supported.md Aevral is an AI security reviewer for GitHub pull requests. GitHub only. Which source files each product reads, what Aevral does not look for, and who it is not for. Aevral is an AI security reviewer for GitHub pull requests. This page states the limits plainly, so you can tell in a minute whether it fits. ## Code hosts - **Supported:** GitHub, through the public [GitHub App](https://github.com/apps/aevral), on personal accounts and organizations. - **Not supported:** GitLab, Bitbucket, Azure DevOps, and self-hosted Git servers. ## Languages These lists come from the product code, not from a marketing matrix. No per-language accuracy figure is published. - **PR security review** reads the changed files of a pull request. When a pull request is large, it reads these source files first: TypeScript, JavaScript, Python, Go, Ruby, Java, Rust, PHP, and Elixir. The review says which files it covered. - **Whole-repo scan** reads only these source files and drops everything else on purpose: TypeScript and JavaScript (including `.mjs`, `.cjs`, Vue, and Svelte files), Python, Go, Ruby, PHP, Java, Kotlin, Rust, C#, Swift, Scala, SQL, and GraphQL. Vendor and build folders such as `node_modules`, `vendor`, and `dist` are skipped. ## What Aevral looks for - **PR security review:** every security-relevant class it can ground in the diff: access control, business logic, SQL and command injection, XSS, SSRF, path traversal, unsafe deserialization, token and session flaws, LLM-integration risks, and configuration and hardening (CORS, cookie flags, security headers, debug exposure, TLS, weak crypto, rate limiting on auth endpoints, resource exhaustion, risky dependency changes, secrets exposed in logs or URLs, infrastructure exposure, removed security controls). Out of scope: style, naming, typing, tests, documentation, and general code quality. Advisory, never blocks a merge, at most five findings per review. - **Whole-repo scan:** authorization, IDOR, and business-logic access control only. A finding is a lead with evidence, not a confirmation. Details: [What a PR review looks like](/docs/what-a-pr-review-looks-like) and [What a scan looks like](/docs/what-a-scan-looks-like). ## What Aevral does not do - Not a secret scanner. PR review flags keys and tokens a pull request adds, and secrets a change writes to logs, errors, URLs, or client bundles; it does not scan history or the whole tree. - Not dependency, SCA, or supply-chain scanning. PR review reads the dependency changes a pull request makes (a new, downgraded, unpinned, or plain-HTTP package, or a removed lockfile integrity hash); it does not look up known vulnerabilities (CVEs) or scan the dependencies you already have. - Not memory-corruption analysis. - Not a general SAST, and not a replacement for one. See [Works alongside](/docs/works-alongside). - PR review does not catch a missing check in a file the pull request did not change. It reads the diff and the changed files, so if a change needs a guard added somewhere else (another endpoint, a sibling function, a log scrubber) and that file is untouched, the review cannot see the gap. - It never applies or merges a fix. Suggested fixes are for you or your coding agent to apply. ## Who it is not for - **Teams with no pull-request workflow.** Aevral reviews GitHub pull requests; if code lands without one, the review has nothing to read. - **Enterprise AppSec buyers who need demos, SSO, and a procurement process first.** Aevral is self-serve. The Enterprise console features (SSO / SAML, RBAC, audit log, multiple GitHub organizations, invoice and PO tooling) are not shipped yet. Enterprise scan volume is by quote: [Pricing and plans](/docs/pricing-and-plans). - **Buyers who need audit evidence** for a certification framework, or control mapping on pull requests. Aevral is a security reviewer; its findings are leads for engineers, not audit evidence. Ask through the [contact form](https://tally.so/r/PdxdbV) and we will point you to the right product from the same company. Open-source maintainers are welcome: public repositories get PR review free (500 reviews per organization per month) and one authorized public-repo scan per calendar month. Questions: [contact form](https://tally.so/r/PdxdbV). --- ## Usage API URL: https://docs.aevral.com/docs/usage-api Markdown: https://docs.aevral.com/docs/usage-api.md GET /v1/usage returns where an Aevral organization stands: PR review plan, usage and cap, scan status, and, for PR review, a link an owner or admin opens to subscribe or upgrade. Read-only. An agent cannot buy or change a plan. Read-only. The response can carry a billing link, but only an owner or admin of the organization can act on it, in the console. An `aevr_` key cannot subscribe, upgrade, change a plan, or raise a spend cap. ## Request ``` GET https://aevral-worker-prod.fly.dev/v1/usage Authorization: Bearer aevr_... ``` - The organization comes from the key alone. Query, body and header org values are ignored for keys. A key never reads another organization. - Use a key that is already in your environment. Coding agents must never ask the human to paste a key into chat, and never put a key in a URL. - Responses carry `Cache-Control: private, max-age=30`. Reads are limited to about 60 a minute per organization and never count against scan quota. ## Response ```json { "org_id": "", "pr": { "state": "free", "plan": null, "used": 25, "limit": 25, "at_cap": true, "resets_at": "2026-10-01T00:00:00.000Z", "trial_ends_at": null, "action": { "kind": "subscribe", "human_action_required": true, "who": "an owner or admin of this organization", "url": "https://app.aevral.com/billing?org=&plan=starter&ref=agent#plans", "suggested_plan": "starter", "plans_url": "https://aevral-worker-prod.fly.dev/v1/pr/billing/plans", "message": "Private pull requests get a neutral Check and no review until 2026-10-01. An owner or admin can upgrade at the link." } }, "scans": { "state": "trial", "plan": null, "used": 1, "limit": 2, "at_cap": false, "resets_at": null, "action": null } } ``` ### `pr` (PR security review) | Field | Type | Meaning | | --- | --- | --- | | `state` | `trial` \| `trial_used_up` \| `free` \| `paid` \| `attention` | `attention` means the subscription is past due or unpaid. | | `plan` | `starter` \| `pro` \| `business` \| `scale` \| null | Set only when `state` is `paid`. | | `used` | integer \| null | Private reviews counted against `limit` this period. Paid: PR reviews, one per pull request per billing period (re-pushes do not count again). Free and trial: each pull request head counted once. Null when there is no meter. | | `limit` | integer \| null | The allowance for this period: the trial ceiling, the free monthly allowance, or the plan's PR reviews. | | `at_cap` | boolean | True when `used` has reached a non-zero `limit`. On a paid plan, the first time over the plan adds an extra quarter of the plan free, once; after that, extras at $0.99 per PR review if the owner chose them. Otherwise, on a private pull request past the cap, the GitHub Check is neutral and no review runs. A neutral Check can also mean payment attention, paused paid extras, or an hourly or daily cap: read its title. | | `resets_at` | ISO timestamp \| null | Free: the first day of next month (UTC). Paid: the end of the current billing period. Trial: null. | | `trial_ends_at` | ISO timestamp \| null | Set only when `state` is `trial`. | | `action` | object \| null | What a human can do next. See below. | ### `pr.action` | Field | Type | Meaning | | --- | --- | --- | | `kind` | `subscribe` \| `upgrade` | `subscribe` for an organization with no paid plan; `upgrade` for a paid plan at its cap with a higher plan available. | | `human_action_required` | `true` | Always true. Nothing is bought until a human confirms in the console. | | `who` | string | Who must open the link: `an owner or admin of this organization`. | | `url` | string | Console Billing page for this organization, with the suggested plan when there is one. The console rechecks live state on arrival and says so if the link no longer applies. | | `suggested_plan` | `starter` \| `pro` \| `business` \| `scale` \| null | The lowest available plan whose included reviews cover twice the current usage, or the highest available plan when none does. Null when no plan is suggested. | | `plans_url` | string | `GET /v1/pr/billing/plans`: the live PR plan ladder. | | `message` | string | One sentence you can show the human. | `action` is null when plans are not on sale, the subscription is past due or unpaid, a checkout is still pending, the paid plan is under its cap, cancelling, already upgrading, or on the highest plan. ### `scans` (whole-repo scans) Status only: `state`, `plan`, `used`, `limit`, `private_limit`, `at_cap`, `resets_at`, and `action`, which is always null. `private_limit` is the private-scan allowance: 0 for an organization with no scan plan and no trial left, which can still scan public repositories. Scan plan changes happen in the console Billing page. | `state` | Meaning | | --- | --- | | `paid` | A scan plan is active. `plan` is `team`, `business` or `scale`; `resets_at` is the end of the billing period. | | `attention` | The scan subscription is past due or unpaid. `plan`, `used`, `limit` and `resets_at` are null. | | `included` | An included monthly allowance with no scan plan. `resets_at` is the first day of next month (UTC). | | `trial` | The 14-day private trial: `limit` is 2, `resets_at` is null. | | `free` | No plan and no trial left: one public-repository scan a month, no private scans (`private_limit` is 0). | ## Prices Quote numbers only from the public plans endpoints, never from memory: - PR review, USD: `GET https://aevral-worker-prod.fly.dev/v1/pr/billing/plans` - Whole-repo scans, USD: `GET https://aevral-worker-prod.fly.dev/v1/billing/plans` Both return `{"plans":[{"id","included","monthly_cents","overage_cents"}]}` and need no key. Amounts are in cents, excluding VAT. The human-readable ladders are on [Pricing and plans](/docs/pricing-and-plans). ## Scan refusals point here When `POST /v1/scans` is refused with 402 (`quota_exhausted`, `overage_confirmation_required` or `spend_cap_reached`), the body adds `usage_url` pointing to `GET /v1/usage`. A spend-cap refusal is not a plan problem: an owner may need to raise the cap in the console. ## Errors | Status | Body | Meaning | | --- | --- | --- | | 401 | `{"error": ...}` | Missing, malformed or invalid key. | | 404 | `{"error":"unknown_org"}` | The key's organization no longer exists. | | 405 | `{"error":"method_not_allowed"}` | Only GET. | | 429 | `{"error":"rate_limited","retry_after":N}` | Too many reads. Wait `retry_after` seconds (also in the `Retry-After` header). | | 502 | `{"error":"usage_unavailable"}` | Usage could not be read. Retry later. | | 503 | `{"error":"usage_unconfigured"}` | The endpoint is not available on this host. | ## For coding agents - Call this endpoint only with an `aevr_` key already in the environment. Never ask the human to paste one into chat. - Show the human `pr.action.url` and say who must open it: an owner or admin of the organization. - Never say you upgraded or subscribed. Never open the link in an automated browser. - Quote prices only from `plans_url`. See also [Console and API keys](/docs/console-and-api-keys) and [For AI agents](/docs/for-ai-agents). --- ## What a PR review looks like URL: https://docs.aevral.com/docs/what-a-pr-review-looks-like Markdown: https://docs.aevral.com/docs/what-a-pr-review-looks-like.md The PR security review as it runs today (free tier, on from install): an advisory Check plus inline comments on added lines, up to five findings, silence when clean. Installing the App starts PR reviews. Setup Complete can still turn them off. An opt-out stays off while the App stays installed; reinstalling the App turns reviews back on. Free includes 25 private reviews per month after the trial, and paid plans are live in the console. Security review, handled. The review is its own product, not one of the included whole-repo scans and not a replacement for the scan plans. Reviews start at install. The owner can turn them off. USD per organization, B2B; paid plans are live in the console. ## How a review runs For the mechanics behind these steps (what is read, how files are packed, the grounding gate, verdicts, and the limits as numbers), see [How Aevral reviews a PR](/docs/how-aevral-reviews-a-pr). 1. **Trigger.** Pull requests opened, reopened, and synchronized after the App is installed. After the first review of a pull request, a re-push reviews only what changed since the last reviewed commit; a force-push, a rebase or a reopen gets a full review again. On paid plans a pull request counts as one PR review per billing period, however often it is re-pushed; on Free and the trial, each new head is a new review and the same head is never counted twice. 2. **Read.** The diff of the changed files is read, most security-relevant first, and the most security-relevant files are also read in full at head. Generated files, lockfiles, and locale and fixture data are skipped by rule, and the review names them; the changed dependency lines of a lockfile are still read for risky dependency changes. Aevral reviews every security-relevant class: access control, business logic, SQL and command injection, XSS, SSRF, path traversal, unsafe deserialization, token and session flaws, LLM-integration risks, and configuration and hardening (CORS, cookie flags, security headers, debug exposure, TLS, weak crypto, rate limiting on auth endpoints, resource exhaustion, risky dependencies, secrets exposed in logs or URLs, infrastructure exposure, and removed security controls). Only non-security topics such as style, naming, typing, tests, and documentation are out of scope. 3. **Post.** An advisory Check on the head commit plus inline comments pinned to the exact added lines. Never blocks a merge. Each finding is a lead, not a confirmed exploit. On a re-push, a new review is posted only when there are new findings; the summary review is edited in place. To have the whole pull request reviewed again, comment `@aevral review` (see [GitHub App permissions](/docs/github-app-permissions)). 4. **Bound.** Up to five findings per review; when there are more, the Check says how many were held back. Likely issues Aevral could not prove are listed under "Unverified, worth a look", at most five, in the Check and in one collapsed Aevral comment on the pull request conversation, edited in place on each push; they are never inline comments and never counted as findings. Silence when there is nothing to publish or list. On a large pull request, Aevral reviews the most security-relevant files first, and the review says which files it covered. ## Grounded on added lines Every inline comment is pinned to a line the pull request actually added, on a file the pull request actually touched. If the evidence cannot be grounded, the comment does not post. A large pull request gets its most security-relevant files reviewed first, with the covered files named, never a partial review sold as complete. A pull request past the reading budget gets a Check that says it was too large to review. ## Fix with your agent Each inline finding comment ends with a collapsed **Fix with your agent** block: a prompt you paste into Claude Code, Cursor, or Codex. It names the file and line, the problem, the missing control, and the suggested fix (a lead, not a patch). It asks your agent to check the finding against the code first, make the smallest change with a test, and stop before commit or push. Text that comes from the reviewed code sits in a marked block the agent is told not to follow. Aevral never applies or merges a change. The Check summary and the review footer also carry a one-line notice: installing Aevral accepts the Terms and DPA for your organization. ## Plans Free tier: public repositories are free, 500 reviews per organization per month, and private repositories get 25 pull-request reviews a month free. Before anyone signs in, 25 private reviews a month are free; connecting the install in the console starts a 14-day private trial with a 500-review cap, no card. Paid plans are purchasable in the console. PR security review plans, USD per organization per month (paid plans live in the console): | Plan | Price | Included private reviews | Past included | | --- | --- | --- | --- | | Free | $0 | 25 private pull-request reviews per month | Subscribe in the console, or wait for the next month. | | Starter | $49 | 100 PR reviews a month, re-pushes free | $0.99 per extra PR review, if the owner chose extras | | Pro | $199 | 500 PR reviews a month, re-pushes free | $0.99 per extra PR review, if the owner chose extras | | Business | $999 | 3,000 PR reviews a month, re-pushes free, plus a monthly export of review results (supporting audit evidence) | $0.99 per extra PR review, if the owner chose extras | | Scale | $1,999 | 6,500 PR reviews a month, re-pushes free, everything in Business | $0.99 per extra PR review, if the owner chose extras | | Enterprise | From $4,000 | By quote, contact us | By quote | On paid plans, a PR review is counted per pull request per billing period; a re-push does not count again. The first time an organization goes over its plan, it gets an extra quarter of the plan's PR reviews free, once. After that, $0.99 per extra PR review if the owner turned extras on, at checkout or later in the console; otherwise reviews pause until the next billing period, with a Check on the pull request explaining why. Extras stop at a monthly spend cap the owner can change in the console. Business and Scale add a monthly export (CSV) of the security reviews Aevral ran on your pull requests, with coverage and results, that may serve as supporting evidence for automated security testing in audits (for example ISO 27001 A.8.29, SOC 2 CC8.1). Scale is the highest self-serve plan (Business for an organization on an older price list); at its limit, the path is extras or contact us about Enterprise, from $4,000 a month. Every paid plan includes the standard [DPA](https://trust.ismscopilot.com/dpa#17-aevral). Existing subscriptions keep their price. On Free and the trial, failed or skipped reviews release their reservations; a new head commit is a new review, while replaying the same head does not count twice. Public reviews remain free, 500 per organization per month, and do not consume the private allowance. Scan and PR allowances are separate. Using reviews as evidence in an audit, including what to show an ISO 27001 auditor and where the record lives on each pull request: https://aevral.com/audit-evidence. ## Permissions note Each review runs on a token capped to Contents read, Metadata read, Checks write and Pull requests write, minted for that job and that repository. Your organization accepts the App permissions at install; the App holds more than the token, reserved for announced features that are not live. Reviews start at install; Setup Complete or the Reviews page can turn them off. See [GitHub App permissions](/docs/github-app-permissions). ## Start Install the [GitHub App](https://github.com/apps/aevral) (reviews start on the next pull request), then [sign in to the console](https://app.aevral.com/login) with GitHub; it connects the matching install automatically. The whole-repo scan is a separate product with its own price: [What a scan looks like](/docs/what-a-scan-looks-like). --- ## What a scan looks like URL: https://docs.aevral.com/docs/what-a-scan-looks-like Markdown: https://docs.aevral.com/docs/what-a-scan-looks-like.md The whole-repo scan: press Scan in the console, get a GitHub Check, a console report, and a suggested fix. The whole-repo scan is the first Aevral product. One scan covers your repository's default branch: cross-file context, authorization rules, the business logic behind who has access to what. ## How a scan runs 1. **Trigger.** You press Scan in the Aevral console. Aevral scans at rest, on your trigger. Not on every push. 2. **Read.** The scan goes through the default-branch snapshot of the repository, following authorization logic across files. 3. **Report.** You get a GitHub Check on the scanned commit plus the full report in the console. Each finding carries its evidence: the code paths and rules behind it. 4. **Suggested fix.** Each finding comes with a suggested fix you apply in Claude Code, Cursor, or Codex. Nothing merges without you. Review every finding before you change production code. A scan is not a clean bill of health. ## What the scan reads Today the scan reads: - **Authorization**: who has access to what, enforced where, missed where. - **IDOR**: object-level access checks that are missing or wrong. - **Business-logic access control**: rules the code intends but does not enforce. The scan does not read injection, XSS, SSRF, path traversal, deserialization, token and session flaws, or LLM-integration risks. Those classes are covered by [pull-request reviews](/docs/what-a-pr-review-looks-like), on the diff only. Not memory corruption. Not secrets, dependencies, or supply chain. Not a general SAST. If you need those classes today, keep the tools that do them: [Works alongside](/docs/works-alongside). ## What you get | Surface | What it carries | | --- | --- | | GitHub Check | The scan outcome on the scanned commit, with a link to the report | | Console report | Every finding with its evidence and the affected code paths | | Findings worklist | Remembered findings across scans for that repository, with a human archive ([the findings worklist](/docs/findings-worklist)). | | Suggested fix | A prompt you apply in Claude Code, Cursor, or Codex. Nothing is applied without you. Aevral does not commit or open pull requests. | ## Plans Scan plans, USD per organization, HT B2B, excluding VAT: - **Public repositories: $0.** 1 authorized public-repo scan per calendar month. Public repositories here means authorized public repositories, not that the Aevral service is open source. Authorization is required to scan; findings are not auto-published. - **Team: $99 / org / month.** 4 default-branch scans per month, then $29 per extra scan. - **Business: $399 / org / month.** 16 default-branch scans per month, then $19 per extra scan. - **Scale: $1,699 / org / month.** 100 default-branch scans per month, then $17 per extra scan. - **Enterprise: get in touch.** How the scan reaches its findings, in prose: [How Aevral hunts](/docs/how-aevral-hunts). Full details and the PR review ladder (free tier live from install; paid plans live in the console): [Pricing and plans](/docs/pricing-and-plans). The canonical marketing source is [aevral.com/pricing](https://aevral.com/pricing). ## Start Install the [GitHub App](https://github.com/apps/aevral), then [sign in to the console](https://app.aevral.com/login) with GitHub; it connects the install automatically. Free organizations get two private scans in 14 days; paid first scans use the included scan allowance. --- ## Works alongside URL: https://docs.aevral.com/docs/works-alongside Markdown: https://docs.aevral.com/docs/works-alongside.md Aevral is a security agent for your code. Claude Security and Codex Security are the named competitors. This page is the rest of the field: SAST, SCA, and general review tools you keep. Aevral is an AI security reviewer for GitHub pull requests. Named competitors: [Claude Security](https://aevral.com/compare/claude-security) and [Codex Security](https://aevral.com/compare/codex-security). This page is not those two. It is the SAST, SCA, and general review tools you already run. Aikido also publishes Code Security Audit and Deep Review for authorization, IDOR, and business-logic reading of source. Keep Aikido's platform. Detail: [aevral.com/compare/aikido](https://aevral.com/compare/aikido). ## The complementary roster | Tool | Its stated job | | --- | --- | | [Semgrep](https://semgrep.dev) | Rule and dataflow matching plus AI-powered detection whose scope includes IDOR and broken authorization | | [Snyk Code](https://snyk.io) | Static analysis (SAST) on your own code | | [CodeQL](https://codeql.github.com) | Deep semantic queries over a code database | | [Aikido](https://www.aikido.dev) | Platform (SAST, dependencies, secrets, IaC, pentest) plus [Code Security Audit](https://www.aikido.dev/code/code-audit) and Deep Review | | GitHub Copilot code review | General code review assistance on pull requests | | [Bugbot](https://cursor.com/bugbot) | AI code review for bugs on pull requests | | [CodeRabbit](https://www.coderabbit.ai) | AI code reviews on pull requests plus CodeRabbit Security monitoring | | [Greptile](https://www.greptile.com) | AI code review with full codebase context; TREX test agent | | [Socket](https://socket.dev) | Supply chain and package risk | | [SonarQube](https://www.sonarsource.com/products/sonarqube/) | Quality and security analysis with quality gates; Hunter Agent for logic flaws | | [ZeroPath](https://zeropath.com) | AI-native SAST platform: business-logic and broken-auth detection, SCA, secrets, IaC, DAST, pull-request reviews with autofix | | [AISLE](https://aisle.com) | AI-native vulnerability management: snapshot scans in cloud, on-prem, or air-gapped deployments, agent analysis, fix agents | | [Tachyon](https://tachyon.so) | AI code security reviews in full-codebase context with exploit validation and a fix per finding | | [Gecko Security](https://gecko.security) | Semantic graph analysis of code, logic, and infrastructure, with cross-repo scanning and a PR review bot | | [Nullify](https://nullify.ai) | Autonomous product security: validated findings driven to merge-ready remediation PRs | Each of these does its stated job. Aevral's job is the security agent: scan the repository, review pull requests, suggested fixes. Open-source models, run in the US today; EU-only processing announced. Aevral's own PR security review is live (reviews start at install; Setup Complete or the Reviews page can turn them off); paid plans are live in the console. ## What Aevral adds - Whole-repo scan and PR review on the same GitHub App. - A suggested fix you apply in Claude Code, Cursor, or Codex. Nothing merges without you. - Open-source models, run in the US today; EU-only processing announced ([aevral.com/security](https://aevral.com/security)). ## Coverage today Today the scan reads authorization, IDOR, and business-logic access control only; pull-request reviews also look for injection, XSS, SSRF, path traversal, deserialization, token and session flaws, and LLM-integration risks on the diff. Not memory corruption. Not secrets, dependencies, or supply chain. Keep the tools that cover those classes until those classes ship on Aevral. Worked examples per tool live on the marketing site: [aevral.com/checks](https://aevral.com/checks). Scan details: [What a scan looks like](/docs/what-a-scan-looks-like). Pricing: [Pricing and plans](/docs/pricing-and-plans).