Code review for the agent era.

Do not send AI into your codebase blind.

DevReadr records what your code is supposed to mean, reviews every pull request against it, and hands your coding agent a structured work order when something is wrong.

Request private beta Connect your agent

Pull request: activate tenant at checkout so users skip the webhook waitChanges requested

TL;DR: 4 findings, 1 of them intent conflict. Worth fixing before merge.

This PR moves tenant activation into createCheckoutSession to skip the webhook wait, but the stored corrected intent explicitly forbids this. Worse than the description implies, activation fires at session creation, before the user has even seen the payment page, so any visitor who starts checkout gets an active workspace with no payment at all.

critical intent_conflict api/stripe/checkout.js:16 createCheckoutSession calls activateTenant, violating recorded intent

high bug api/stripe/checkout.js:11 success_url claims activated=1 but redirect success does not prove payment

Real repository. Real review. The diff looked fine; the recorded intent caught it.

A diff tells you what changed. It cannot tell you what the code was supposed to mean.

Record meaning

Readr notes lock what each file is supposed to mean, in plain English. The diff can lie about intent. The record cannot.

No raw source is stored. The database schema rejects it.

Review every pull request against it

A change that contradicts recorded intent is an intent conflict, and it blocks the merge. Reviews run on the newest frontier models with automatic failover.

Or trigger one yourself: comment @devreadr review on any PR. That is the whole integration.

Autopilot hands your agent the fix

Findings become a structured work order: file anchors, instructions, tests, rollback, constraints. Your agent executes it, opens a draft PR, and reports back.

Humans approve the merge. That part is permanent.

Made for agents

Your agent is the first-class user.

Cursor, Claude Code, Windsurf, or your own agent. One key, five tools. Your agent reads recorded intent before it edits, runs reviews, and works Autopilot orders, all from the editor it already lives in.

Every PR comment carries a machine-readable payload. No scraping prose.
Work orders are versioned contracts with anchors, tests, and rollback.
The full contract is public: /api/devreadr/agent-manifest

One snippet connects any MCP client

{
  "mcpServers": {
    "devreadr": {
      "type": "http",
      "url": "https://devreadr.com/api/devreadr/mcp",
      "headers": { "Authorization": "Bearer dvr_<your key>" }
    }
  }
}

Keys are created in the workbench, shown once, revocable any time.

No raw source retained. Ever.

DevReadr stores derived meaning, never your code. The database schema physically rejects raw source. Actions are draft-only until a human approves, and customers control which repositories connect.

Terms Privacy

Connect a repo. Run a review. Decide for yourself.

Founding beta is for the first teams that want their codebase reviewed against recorded intent, with agents doing the work and humans keeping approval.

Founding Beta

$28 first month

Private beta checkout for first cohort.

Starter

Target under $100/month

Multiple codebases for founders and small teams.

Team and Enterprise

Scoped manually

Collaboration, permission review, and guided rollout.

Founding beta launch

See your first intent conflict.

Prefer to collaborate first? Email hello@devreadr.com.