Mined, not written.
Tently reads your merged pull requests, past review threads and docs like ADRs, READMEs and CLAUDE.md, then proposes the decisions your team already made.
Tently reviews every pull request against the decisions your team already made and the callers the diff never touched, and answers Claude Code, Cursor and Codex while they write.
1,522 developers on the waitlist · Already have access? Sign in

Install the GitHub App once. Tently learns what your team decided, then holds every change to it: your coding agent asks while it writes, and the pull request check enforces it before anything merges.
Tently reads your merged pull requests, past review threads and docs like ADRs, READMEs and CLAUDE.md, then proposes the decisions your team already made.
Gate one. Your coding agent asks Tently over MCP before it changes a file, and checks the impact of a change on your working tree, which never leaves your machine.
On every pull request Tently builds a code graph from a fresh clone and follows each changed function, type and export to the code that calls it, inside the diff or not.
Gate two. Verified findings land on the pull request as a check run and inline comments, evidence attached. Make the check required and a regression can’t merge.
New decisions from merged work come back as proposals for your team to approve, edit or reject. A pull request still under review never creates one.
Regressions
A signature changes in one file and three callers break in three others. Tently follows each changed symbol through the code graph to its callers and entry points, then reads them to decide what really breaks.
Watch it catch #214Every changed function, type and export, traced to the code that calls it.
A finding names each impacted symbol and file, or links the decision it breaks.
A change that goes against a decision cites it, with a link to where your team made it.
No callers in the graph? Tently still searches the code and reads likely consumers.
Regression or drift, critical to info, with a confidence score on every finding.
Conventions worth following. A finding that cites only a pattern is capped at low severity.
The product
A decision store your team approves, a gate in your coding agent, a check on every pull request, and findings that earn their place.

Each decision has a rationale, a scope of paths and a link to its source. Approve, edit or reject; approved decisions become what every review checks.
npx @tently/mcp init connects Claude Code or Cursor. Impact checks run on your working tree, so the source never leaves your machine.
A check run and inline comments on every pull request. Make it required and a regression can’t merge.
A stronger model confirms each finding before it posts. Your feedback tunes it, and dismissed stays dismissed.
Keep CodeRabbit, Copilot or cubic for the diff. Tently covers what lives outside it.
Each review clones your repository into a sandbox with an installation token that expires within the hour. When the review ends, the sandbox is destroyed. What stays is decisions, findings and feedback.
Each review runs on a fresh clone in a sandbox that is destroyed when the review ends.
Installation tokens expire within the hour. Tently writes only PR comments and check runs.
Every query is scoped to your organization by Postgres row-level security.
What persists is the decision store, findings and feedback. Never your files.
The film · 1:43
A pull request, checked against the decisions it was about to break.
Pricing
Plans include pull requests reviewed each month. Seats, repositories and the MCP gate are unlimited on every plan.
How PRs are countedTry Tently on your own repos.
$0 /month
30 PRs a month, then reviews pause
Request accessA team on a shared codebase.
$99 /month
250 PRs a month, then $0.50 each if you turn it on
Request accessOrg-wide institutional memory.
$399 /month
1,000 PRs a month, then $0.40 each if you turn it on
Request accessAlready in? Pick a plan on the Billing page of your dashboard. Annual billing or self-hosting? Write to [email protected].
Anything else? Write to [email protected].
No. Tently looks for regressions and decision drift, the problems diff-only reviewers miss. Keep CodeRabbit, GitHub Copilot or cubic for what’s inside the diff; Tently runs next to them.
No. Tently mines decisions from your merged pull requests, review threads and docs such as ADRs, READMEs and CLAUDE.md files. Your team approves, edits or rejects each proposal. You can also write one by hand or import it from a document.
No. Each job clones the repository with a short-lived installation token into a temporary workspace, which is deleted when the job finishes. Tently keeps decisions, findings, embeddings of decisions, repository metadata and your team’s feedback.
Pull request comments and check runs. Nothing else.
Yes, if you make the Tently check required in your branch protection rules. A pull request with a regression then can’t merge until it’s fixed.
Any MCP client. Run npx @tently/mcp init in your repository with an API key from the dashboard: it sets up Claude Code and Cursor directly, installs a skill that says when to call search_decisions and query_impact_zone, and backs up any file before it changes it. Codex and other MCP clients connect to the same server.
By default, requests go through OpenRouter to Anthropic. A workspace can add its own OpenRouter API key instead. Tently doesn’t train models on your code, decisions or findings.
Tently is in early access for teams on GitHub. Join the waitlist with your work email and we’ll be in touch.
Tently is in early access for teams on GitHub. Join 1,522 developers on the waitlist.
# once you’re in: connect your coding agent $ npx @tently/mcp init ✓ registers the Tently MCP server ✓ installs a skill for search_decisions and query_impact_zone ✓ backs up any file before it changes it