Redocly Writer Agent watches every signal your users leave behind, investigates what's wrong, stale, or missing, fact-checks the fix against your code, and opens a pull request with it. You review and merge.
Opens draft PRs, never merges them
Fact-checked against your code
Follows your repo's rules
Adjusts the PR to your comments
Pull request opened
Scheduled run · just now
Evidence
11 searches for "rotate API key" returned no results
4 AI Assistant conversations asked which scope key rotation needs
1 feedback says the setup guide doesn't name the scope
Document the keys:rotate scope in the API key setup guide
Docs rot between edits. And the signals die unnoticed.
Feedback, dead-end searches, and repeated AI questions all point to gaps in your docs. Each one needs someone to notice it and write the fix, so most go unanswered.
How it works
It watches. It investigates. It opens a PR. You decide.
01
Watches your signals
Feedback, search, AI Assistant, page views, support chats, and your own sources over MCP.
02
Investigates with confidence
Connects the evidence to your real docs and acts only on a concrete gap. Otherwise, it does nothing.
03
Opens a pull request
With the fix, the evidence behind it, and passing checks, including a fact check against your code. It follows your PR template and pings your team in Slack.
04
You review & merge
Your normal workflow. Leave a comment or suggestion, and the agent picks it up to refine the fix.
Signals
It learns from how people use your docs
Built-in sources today, with room to plug in your own.
Feedback
Comments and ratings from readers and AI agents. New feedback can start a run right away.
Search
Including searches that return nothing.
AI Assistant conversations
Answers your readers disliked, and their recent questions.
Page views
Shows which pages readers land on, so fixes go where the traffic is.
Support chats
The questions your team answers over and over in Front, connected with our team.
Bring your own – via MCP
Connect your own systems – codebase, support inbox, internal wiki, observability – by registering an MCP server, and the agent uses them as context too.
Your codebase
Coming soon
The agent starts a run when your source code changes and finds where the docs no longer match the code.
Anatomy of an agent PR
Every change shows its work.
No black box. Each pull request shows the exact signals that triggered it, the change itself, and the checks it passed.
It reads like a great teammate's PR: here's what I saw, here's why, here's the change, here's how I checked it.
Document the keys:rotate scope in the API key setup guide
11 searches for "rotate API key" returned no results
4 AI Assistant conversations asked which scope key rotation needs
Feedback: "The setup guide doesn't say which scope the token needs"
What it changed
docs/api-keys.md
To rotate a key, open Settings → API keys.
-Any token can rotate a key.
+The token must have the `keys:rotate` scope.
+See [Token scopes](./tokens.md#scopes) for the full list.
docs/tokens.md · adds an anchor for the scope table
How it checked
Style lint
Redaction check
Fact check against your source code
1 claim flagged for your review: “Rotating a key invalidates the old secret immediately” has no source in your code.
Nothing ships without you
Autonomy without the risk
Every powerful capability is paired with a control that keeps you in charge.
Fully auditable
Every PR cites the exact signals and investigation that justified it.
Conservative by design
Acts only on a concrete gap backed by evidence – and is happy to do nothing. It remembers why you closed a PR and checks new work against it.
The PR is the boundary
No pushes to your default branch. No auto-merge. Ever.
Your data, respected
Never reads secret files, redacts secrets before the model sees them, and blocks a PR that leaks a redacted value.
You're in control
Enable per project, pick its duties, add sources and rules, turn it off anytime.
Capped
Coming soon
A few PRs per project per week. You set the number.
Roadmap
Available now, and on the way
We label this honestly so you always know what you can turn on today.
Available now
Scheduled runs that survey your signals and open PRs
Instant runs on new feedback, from readers and AI agents
When it acts, it opens a reviewable PR with evidence + validation
Fact-checks every change against your linked source repositories
Follows your repo's agent rules (CLAUDE.md, AGENTS.md, Cursor and others)
Remote content integration
Reacts to your comments and adjusts the PR
Bring your own context via MCP
On the way
Instant runs on search, AI Assistant and support signals, not only on a schedule
Starts a run on every code change you merge
GitLab, Bitbucket and Azure DevOps support
Two-way Slack collaboration
Custom duties prepared by you
Frequently asked questions
Writer Agent is free during early access. Access is by request: leave your email and we will get in touch.
No. It opens draft pull requests and never merges them or pushes to your default branch. Your team reviews every change, and the agent refines the fix when you leave a comment.
It checks each change against the source repositories you link and follows your repo's agent rules (CLAUDE.md, AGENTS.md, Cursor and others). If it cannot verify a claim, it flags the claim in the pull request for your review.
GitHub during early access. GitLab, Bitbucket, and Azure DevOps are on the roadmap.
Once we enable early access for your organization, open your project settings in Reunite, turn the agent on, and pick its duties. Then sit back and watch it work.
Each run happens in an isolated sandbox. The agent never reads secret files, redacts secrets before the model sees them, and blocks a pull request that leaks a redacted value.
It finds the problem. You approve the fix.
Now in early access
Let your docs improve from how people actually use them, while your team keeps the final say.
Free and GitHub-only during early access · Talk to us