Your AI agents,
on rails.
Point Worca at a project and describe the work. A deterministic engine drives AI agents through Plan → Refine → Implement → Review — looping until the work clears quality gates, pausing to ask you the questions that matter, and keeping every run isolated in its own git worktree and branch.
Describe the work.
Approve the result.
Agents do the creative work; the engine does the control flow. Every step, verdict, and artifact is recorded — nothing depends on an agent remembering to behave.
Hidden decisions become multiple-choice questions before any code is written. Your answers travel with the plan.
The planner explores the codebase and writes an implementation plan with concrete code snippets.
The refiner reviews and rewrites the plan — v2, v3, … — until no critical or major issues remain.
The implementer follows the approved plan with no deviation, using TDD: red, green, refactor.
The reviewer reads the diff and hands blocking findings back to the implementer — looping until clean.
Loops gate to you past their cycle cap: approve another cycle, or continue with the open issues shown. A finished run ends on its own branch — one click from a PR.
- ?Real decisions, surfaced early2–4 options plus free text — instead of silent assumptions buried in a diff.
- →Answers stay on the recordThe clarify Q&A is appended to the plan, so refiners and reviewers see exactly what you chose.
- ⌂Answer anywhereIn the browser — or straight from Telegram, Slack, Discord, or Teams.
Clarify — the pipeline asks before it assumes.
Every run keeps
its receipts.
The diff per file, per-step costs and durations, the clarify Q&A, agent transcripts, and logs — kept for every finished run, across every project on your machine.
Run detail — diff, per-step costs, one-click PR.
- ±Diff-first reviewFiles changed with +/− counts, ready to open as a pull request from the History view.
- $Costs per stepWhat each agent spent, and how long it took — for every cycle of every loop.
- ≡Transcripts includedFull agent transcripts and logs, filterable by source, level, step, and cycle.
- ⌫Nothing in your repoRun state lives in one SQLite database in your home directory — your checkout stays clean.
Compose your
own pipeline.
Drag agents onto a canvas to build sequential steps, parallel groups, and feedback loops — an agent that emits a verdict can loop back to an earlier step until it passes or hits its cycle cap. Saved workflows appear in the New Pipeline picker.
- ⇄Feedback loops, boundedVerdict-emitting agents loop back with configurable cycle caps — no runaway rework.
- ∥Parallel groupsFan work out across agents where order doesn't matter; the engine joins the results.
- ↺Reset to defaultOne click redraws the standard Plan → Refine → Implement → Review.
Workflow Composer — steps, groups, and loops on a canvas.
Eleven agents.
Zero hardcoding.
Planner, refiner, plan reviewer, implementer, code reviewer, clarify, decomposer, manual-test checklist, web-UI tester, workspace scanner, workspace reviewer. Each agent is a markdown prompt plus a metadata sidecar — new agents drop in without engine changes.
Agents — data-driven, editable, per-agent model & effort.
- ✎AI-assisted agent creationDescribe a new agent and Worca generates its system prompt and wiring — edit, regenerate, save.
- ◎Model & effort per agentPick model and reasoning effort per agent, per workflow, or per run — with a clear resolution order.
- ＋Bring your own modelsRegister any model id — a proxy, a fine-tune, another provider — with per-model routing environment.
Autonomy, with
a perimeter.
Named policy sets, selected per run. Policies compile to hard permission denies on every agent spawn — repo settings can't undo them.
Let agents move fast in sandboxes and throwaway projects.
Protects credential files and blocks publication commands.
Adds environment scrub on spawn, network-egress and cloud-CLI denies, and home-dir credential protection.
- ⛨Your own policy setsStart from a tier, tune the rules, save it under a name, pick it per run.
Guardrails — pick the perimeter per run.
Drive runs from
your pocket.
Bundled two-way channels: get notified on questions, finishes, failures, and cost pauses — and answer back with commands, with allowlist-based authorization.
Spend you can
see — and cap.
Per-run and per-step cost estimates, a Statistics view with spend, time, and outcome charts, and a spend indicator always in the sidebar.
Statistics — spend, time worked, outcomes per day, week, or month.
- ⏸Per-pipeline cost capA runaway run pauses itself — resumable only with an explicit override.
- ■Total budgetPauses everything and blocks new runs until the weekly or monthly window resets.
- ✓Estimates everywherePer run, per step, per agent — visible while the run is still going.
The rest of
the toolbox.
One install.
Three ways in.
Worca ships as a web UI, a CLI, and an installable skill for your coding agent — all running the same deterministic engine. Requires Node.js ≥ 22.13.
One npm package: engine, web UI, and CLI included.
npm install -g @worca/app
Add a project, click New pipeline, describe the task.
worca --ui
Point at any project and describe the work.
worca --project . \
--prompt "Add a /search endpoint"
Want a zero-cost tour first? --mock runs the full pipeline offline — no tokens spent. See the full quick start.
Ship work you
didn't babysit.
Worca is open source and ready for your next project.