Greptile

Reviews run in GitHub and GitLab across more than thirty languages, including self-hosted GitLab behind a reverse proxy; context is pulled from Jira, Notion, Google Drive and Sentry; findings are delivered into Slack and the IDE; and Fix with your Agent sends a comment with file, line numbers and suggested fix straight to Claude Code, OpenAI Codex, Conductor, Cursor or Devin through a local bridge CLI, with a Fix All control sending every issue at once and comments auto-marked as addressed when a commit touches the flagged files. A hosted MCP server, REST API, npm CLI and Zapier integration extend the surface further, and the Partner Program adds maintained context for OpenAI, Vercel, Stripe, PostHog, Datadog, WorkOS, Mintlify and Braintrust.

Sourcegreptile.com/docs/integrations/fix-with-your-agent and greptile.com/changelog 2026-04-08read 2026-08-30

The v5 review engine runs a swarm of narrowly scoped agents in parallel, each exploring one hypothesis for a potential bug, with results aggregated into a single review; the vendor reports median review time falling from 5:04 to 2:25 and author-addressed comments rising from 52 to 66 percent in production A/B tests. The agent expands beyond the diff to find similar logic, traces dependencies, reads git history and runs iterative passes that challenge their own hypotheses, with the maximum number of agentic turns configurable per review; TREX runs as a companion agent writing and executing tests, and the security reviewer chains Opengrep pattern scanning, dependency analysis and AI exploit detection.

Sourcegreptile.com/blog/greptile-v5 and greptile.com/changelog 2026-08-05read 2026-08-30

A graph index of the entire repository maps how functions, classes, files and directories connect and updates continuously, so reviews reason about ripple effects beyond the diff; Repo Clusters and a context.repos field give read-only access to related repositories during a review, up to seven per review, with clusters suggested from shared contributors over the previous 90 days; files.json points reviews at existing schemas, API specs and architecture docs in the repository; context is pulled from Jira, Notion, Google Drive and Sentry; the Partner Program supplies partner-maintained documentation for supported third-party APIs; and imported AI rules files including CLAUDE.md and Cursor rules become org-wide context. Developers query the codebase in natural language and receive answers referencing specific files, functions and classes.

Sourcegreptile.com/docs/code-review/cross-repo-context and greptile.com/changelog 2026-06-02 and 2026-04-15read 2026-08-30

Greptile posts its reviews and, where enabled, approves pull requests on its own under a customer set risk ceiling with exclusion filters and a never approve list covering authentication, secrets, billing, migrations, infrastructure, CI and public APIs, and custom rules shape what it flags. These are customer controlled constraints on the agent, and no step where a person approves the agent's action before it lands is documented.

Sourcegreptile.com/docs/code-review/first-pr-review and greptile.com/llms.txtread 2026-09-29

SOC 2 Type II certified with independent audits and data encrypted at rest and in transit, single sign-on with SAML documented as its own page in the self-hosting section, audit logs available to enterprise, and a dedicated security section covering compliance practices; the whole service self-hosts into a customer VPC or an air-gapped environment with customer-supplied LLM providers so code, secrets and inference never leave the network, and each sandbox and integration is scoped through the customer's own GitHub or GitLab installation. No trust portal, named audit firm or published attestation report has been found.

Sourcegreptile.com/security, greptile.com/docs/security/selfhost and greptile.com/docs/deployment-optionsread 2026-08-30

An analytics dashboard tracks pull requests reviewed, addressed rate, critical bugs caught, merge times and upvote to downvote ratios, filterable by team, repository, author and time period with export; each inline comment carries a confidence score and a P0 to P2 severity badge, review summaries include auto-generated sequence diagrams of call flows, and the review footer shows a review counter and a link to the last reviewed commit; enterprise deployments have audit logs, and TREX attaches execution evidence to failing findings. These report on review outcomes and team throughput rather than retaining a per-action trace of the agent's own reasoning, and no session record, execution replay or agent run history is documented.

Sourcegreptile.com/docs/analytics and greptile.com/changelog 2026-04-15read 2026-08-30

Memory and Learning is a named system that learns team preferences from PR comments, replies to Greptile, reactions and whether comments were addressed, and turns repeated patterns into suggested rules. The page states no scope boundary, lifetime or way to review and delete what was learned, so it meets Partial rather than Full.

Sourcegreptile.com/docs/how-greptile-works/memory-and-learningread 2026-09-29

The entire service can be self-hosted, including in air-gapped environments, by Docker Compose on a single Linux VM for up to 100 developers or Kubernetes for 100 or more with horizontal scaling and high availability, with published sizing requirements and a documented migration path between the two via parallel deployments; an AWS Terraform module provisions VPC, EC2, RDS PostgreSQL and ElastiCache Redis and bootstraps the application, and manual setup covers GCP, Azure and on-premises.

The deployment repository is public at github.com/greptileai/akupara, self-hosting requires a license, custom LLM providers are supported so inference stays on customer infrastructure, and self-hosted GitLab behind a reverse proxy is supported. Cloud remains available for teams wanting no infrastructure management.

Sourcegreptile.com/docs/deployment-options and greptile.com/docs/security/selfhostread 2026-08-30

The Partner Program supplies vendor-curated review context for third-party APIs, so a pull request touching a supported partner API, SDK, config or integration is reviewed against partner-maintained documentation, implementation rules and common failure patterns, with comments marked by partner logo and linked back to the relevant docs; it is enabled by default for all users with featured partners including OpenAI, Vercel, Stripe, PostHog, Datadog, WorkOS, Mintlify and Braintrust.

Greptile additionally ships two named agents, the code reviewer and TREX, and imports existing AI rules files including CLAUDE.md, AGENTS.md and Cursor rules as org-wide context. No browsable catalog, marketplace or gallery of installable agents or templates is documented.

Sourcegreptile.com/partners and greptile.com/changelog 2026-06-22read 2026-08-30

Reviews start automatically when a pull request opens in GitHub or GitLab and re-run on later commits under configured review triggers, so work reaches the agent with no person starting it; it can also be invoked by @greptileai, a re-trigger control, a reply in the thread, or the CLI on a local branch.

Sourcegreptile.com/docs/quickstart, greptile.com/llms.txt and greptile.com/cliread 2026-09-29

Review settings let the customer choose which AI model powers reviews and set the maximum number of agentic turns per review, balancing speed, depth and cost; Model Inversion detects from commit trails, branch prefixes and PR titles when a pull request was authored by a coding agent and routes the review to a different model, so Claude-authored code is reviewed by GPT and GPT-authored code by Claude; and self-hosted deployments support custom LLM providers so models run entirely on customer infrastructure.

Sourcegreptile.com/changelog entries of 2025-09-26 and 2026-07-22, and greptile.com/docs/deployment-optionsread 2026-08-30

A documented REST API exposes Index Repository, Query Repos, Search Repos and Get Repository Info alongside webhooks and API details, with API keys issued from app.greptile.com/settings/api; a hosted MCP server at api.greptile.com/mcp is configured by bearer token in Claude Code, Claude Desktop, Cursor and VS Code and lets assistants trigger and re-run reviews, inspect results, manage custom context and update repository rules; an npm CLI runs reviews locally with --json for machine-readable output and --agent for agent workflows and scripts; a Greptile plugin ships in Anthropic's official Claude Code marketplace; and a Zapier integration is published.

Sourcegreptile.com/docs/mcp/setup, greptile.com/changelog 2025-09-15 and 2026-06-04read 2026-08-30

Testing the customer's change is the product: every pull request is reviewed for logic, security and cross file bugs with confidence scores and P0 to P2 severity, and TREX writes targeted tests for the change and runs them in a sandbox against the repository's own services, attaching logs and traces, which is a quality gate in the release path.

Sourcegreptile.com/trex and greptile.com/llms.txtread 2026-09-29

All documented action runs through programmatic interfaces: the GitHub and GitLab APIs for reviews, comments and approvals, a graph index of the repository, an isolated sandbox in which TREX executes generated tests against the repository's existing services and framework, a CLI, an MCP server and REST endpoints. TREX attaches logs, screenshots, traces, scripts or videos as evidence when a test fails, but no browser tool, computer-use capability or operation of software lacking a programmatic interface is documented on the TREX page, the agent page or in the changelog.

Sourcegreptile.com/trex and greptile.com/changelog 2026-06-15read 2026-08-30