Features
Built for teams that care what ships
Every part of Reqursor Development exists to answer one question before anything reaches your repository: is this change actually done?
The pipeline
Every stage gets a fresh session
Each stage starts a new agent session, and stages hand work to each other as files, not as one long conversation that drifts. That is the main defense against context rot.
01
Implement
The agent writes the change in your repository. Optionally it runs several candidates and keeps the one that passes the most checks.
02
Self-review
A quick pass to tidy up the diff before anything is judged.
03
Checks
Your build, lint, typecheck and tests, plus built-in scans. No AI involved, just pass or fail.
04
Review
An AI reviewer returns a structured verdict, citing file and line evidence for every acceptance criterion.
05
Verify
A fresh, independent agent sees only the diff and the ticket, never the implementer’s reasoning, and re-derives whether the work is done.
06
Security
A security scan of the change. Low-risk findings are fixed automatically and checked again.
07
Fix loop
Failing checks or review findings get a limited number of fix attempts, each one checked again from scratch.
08
Ship decision
It ships only if the verdict says ship, every required criterion has evidence, every check is green and no budget was breached.
Checks
Your checks are the ship floor
Checks run your own commands and built-in scans. They don’t use AI, and any failing check vetoes the ship, no matter what the reviewer thinks.
Build
Your build command must succeed.
Typecheck
No new type errors.
Lint
Your linter, your rules.
Tests
Your test suite, run on every change.
Secret scan
Blocks API keys, tokens and passwords before they’re committed.
Stub scan
Catches TODOs, stubs and placeholder code passed off as done.
Dependency policy
Block, warn on, or require approval for new packages.
Diff coverage
New and changed lines must be covered by tests (optional).
Diff-size budget
Keeps each change small enough to review.
Test-first
Flags changes that arrive without tests.
Runtime smoke test
Boots your app and checks a health URL before shipping (optional).
Custom commands
Any command you can run in a terminal becomes a check.
- Your base branch must pass before a run starts. The agent never builds on a broken foundation
- Flaky checks can be retried with extra settings before they count as failed
- Missing toolchains are reported as environment problems, never “fixed” by editing code
The queue
One queue for all your work
Write tickets in the built-in queue or import them from your tracker. The queue is the single place the agent takes work from.
- Board columns: Draft, Ready, Queued, Running, Needs review, Done
- An acceptance-criteria checklist is required before a ticket can be Ready
- Labels, priority and dependencies decide the run order
- Import from GitHub, GitLab, Jira and Linear; results are posted back when a ticket ships
- Run one ticket, run all Ready tickets, or switch on Autopilot per project
- Preview the run order and estimate the cost before you start
Try it: drag a Ready ticket into Running, or switch on Autopilot.
Press Space or Enter, or click, to pick up a ticket. Move it with the arrow keys or a Move here button, then press Space or Enter, or click the ticket, to put it down. Escape cancels.
Live run view
See every step, and the proof
Open any ticket to follow the run as it happens, then check the verdict line by line.
- A stage stepper with live status
- The agent’s activity streamed as it works, with secrets redacted
- Check results with their output
- Every acceptance criterion linked to the exact lines in the diff
- Cost, tokens and duration for every attempt
- Implement6m 12s
- Checks1m 48s
- Review58s
- VerifyRunning
- SecurityQueued
- ShipQueued
- Invoices page has an "Export CSV" buttonsrc/invoices/page.tsx:88–96
- Export includes date, customer, amount and statussrc/invoices/export.ts:12–41
- Amounts use the account currencysrc/invoices/export.ts:27
- Tests cover empty and large exportssrc/invoices/export.test.ts:8–64
Human in the loop
You stay in charge of what lands
The agent does the work; your team keeps the final say. Nothing is ever committed silently after a failure.
Needs-review inbox
Tickets that need a decision collect in one inbox, with a badge in the navigation.
Approve & commit
Large or sensitive diffs wait until someone on your team approves them.
Retry with a hint
Send a ticket back with a note for the agent’s next attempt, or skip it.
Escalation reports
When a ticket can’t ship, you get a clear report of what was tried and why it stopped.
Revert
Undo a shipped ticket with a new commit, never a history rewrite.
Post-mortems
Download the full trace of any run as Markdown or JSON.
Cost control
Predictable AI spend
You connect your own Anthropic API key, so you pay Anthropic directly for exactly what you use, and you set the limits yourself.
- Test the connection and choose from the models your key can use
- Pick the model and effort for each pipeline stage
- Caps per ticket, per day and per install, plus limits on run time and tool calls
- Work stops as soon as a cap is hit. It never keeps spending past it
- Budget forecast and cost split by stage
Enterprise? Reqursor Development also runs on Amazon Bedrock, Google Vertex AI or Microsoft Foundry.
- Implement58%
- Review17%
- Verify15%
- Security10%
Observability
Know how it’s doing, without asking it
Every action is recorded, and the numbers that matter are one click away.
Overview
What’s running, what’s queued, ship rate, spend against budget and base-branch health.
Insights
Ship rate, check health, cost trend, token burn, velocity and budget forecast.
Drift canaries
Synthetic tickets that catch a silent drop in model quality before your real tickets do.
Alerts and digest
Threshold alerts, ship-rate regression alerts and a daily digest by email, Slack or webhook.
Audit log
An append-only record of every action, searchable with filters.
Prometheus metrics
A token-protected metrics endpoint to scrape straight into Grafana.
Operations
Built to run unattended
A guided setup takes you from install to your first ticket, and the maintenance work is a button away.
- 1License
- 2Owner account
- 3Connect Claude
- 4Connect repository
- 5Checks
- 6Notifications
- 7First ticket
- Users and roles: Owner, Admin and Developer, plus free Viewers
- Scoped API tokens for automation
- One-click health check and diagnostics bundle
- Updates with automatic rollback if anything fails
- Backups you can download and restore
- Resumes cleanly after a restart or a power cut
How it compares
Not another coding assistant
In-editor assistants help a developer type faster. Cloud agents run on someone else’s machines. Reqursor Development works through your queue, on your hardware, and proves its work.
| Capability | Reqursor Development | In-editor AI assistants | Cloud coding agents |
|---|---|---|---|
| Works unattended from a ticket queue | Built in | Not offered | Built in |
| Your own checks must pass before anything ships | Built in | Varies by product | Varies by product |
| Evidence for every acceptance criterion | Built in | Not offered | Varies by product |
| Independent second review in a fresh session | Built in | Not offered | Varies by product |
| Runs on your own hardware | Built in | Varies by product | Not offered |
| Hard budget caps per ticket and per day | Built in | Varies by product | Varies by product |
| Git guardrails enforced in code | Built in | Varies by product | Varies by product |
Built in Built inVaries by product Varies by productNot offered Not offered
See it on your own codebase
Start a 14-day trial on your own repository, no card needed.