How it works
From a sentence to production
cyql ≈ cycle. Every task runs the same loop, and the loop is durable — it resumes after any failure and replans when the work needs another pass.
Migrate auth to JWT refresh tokens
acme/api · base main · updated 2m ago
- Draft
- Planning
- Awaiting approval
- Dispatching
- Coding
- Creating PR
- Awaiting review
- Reviewing
- Merging
- Deploying
- Shipped
Approach
Introduce rotating refresh tokens backed by a new token store, dual-write during rollout, then switch the client to the refresh flow behind a flag.
Add refresh-token table + migration
migration · M · 2 files
Issue & rotate refresh tokens in auth service
backend · L · 5 files
Switch the web client to the refresh flow
frontend · M · 4 files
The loop
Five stages, repeated until shipped
- 01
Describe the task
Write what you want in plain language, link an issue, or send it from Slack. cyql reads your repo, conventions, and past work to ground the request.
- 02
Plan & decompose
The planner breaks the task into a dependency-aware DAG of subtasks, surfaces key decisions and risks, and waits for your approval when it matters.
- 03
Dispatch the fleet
The orchestrator schedules one isolated pod per subtask — scaling out and back in with the queue. No shared state, no blast radius.
- 04
Code & self-review
Agents write code, run your tests and linters, and review each other's diffs against your standards. Failures route back to a replan, not to you.
- 05
Ship a clean PR
You get a focused, reviewed pull request with a written rationale and a full execution trace. Approve, request changes, or let it auto-merge.
When review can't converge, the task loops back to plan instead of shipping something wrong.
You stay in control
Autonomy you can audit
cyql does the work; you make the calls that matter and can replay every decision.
cyql handles
- ›Reading your repo, conventions, and history
- ›Decomposing the task into a dependency graph
- ›Scheduling and running isolated agents
- ›Writing code, running tests and linters
- ›Reviewing diffs and revising on failure
- ›Opening a clean, well-described PR
you decide
- ›Approve the plan before any code is written
- ›Set guardrails, budgets, and model routing
- ›Review (or auto-merge) the final pull request
- ›Replay the full execution trace, step by step
- ›Turn recurring failures into one-line rules
- ›Choose what each project can and can't touch
Ship your backlog while you sleep.
Start free in minutes. Connect a repo, describe a task, and watch a reviewed pull request show up.
No credit card · Free forever for individuals