# Worca – Deterministic Multi-Agent Pipeline

> Source: <https://worca.dev/>
> Published: 2026-10-24 07:57:48+00:00

# Your AI agents,

on rails.

Describe the work, get back a **reviewed pull request**. A **deterministic engine** runs the agents, loops until the work passes its quality gates, and stops only to ask **you** what matters.

## Describe the work.

Approve the result.

Agents do the creative work; the engine does the control flow. Pick a workflow, or let **Auto** build one for the task. Every step, verdict and file is recorded.

**Clarify**

Hidden decisions become questions before any code is written.

**Plan**

The planner explores the codebase and writes a concrete plan.

**Refine**

The refiner rewrites the plan until no major issues remain.

**Implement**

The implementer follows the approved plan, test first.

**Review**

The reviewer sends blocking findings back until the diff is clean.

A finished run ends on its own branch, **one click from a pull request**. [See the whole pipeline →](https://worca.dev/how-it-works/)

## Everything

a run needs.

Any model, schedules, a record of every run, limits you set outside the agent, and a whole team working in one place.

[How it works](https://worca.dev/how-it-works/)

**Workflows you can see and change.**

Auto picks the workflow, or build one in the Composer: agents, script cards and bounded loops. Every step files what it wrote.

The pipeline →
[Models & costs](https://worca.dev/models/)

**Any model. Every dollar accounted.**

Claude, GitHub Copilot, Ollama or LM Studio on your desk, free OpenRouter models. Caps, budgets, and the hours each run saved.

Models & costs →
[Automation](https://worca.dev/automation/)

**Runs that start while you sleep.**

Schedule a run for tonight or every weekday, chain runs onto each other's branches, and answer from Slack or Telegram.

Automation →
[Ask Worca](https://worca.dev/ask-worca/)

**An assistant that knows your runs.**

Ask what the reviewer flagged, attach a screenshot, let it read the docs page, or have it schedule the next run.

Ask Worca →
[Teams](https://worca.dev/teams/)

**One Worca for the whole team.**

Metrics and a delivery timeline, a team policy kept in git, a shared server behind sign-in, and each person pays with their own key.

Teams →
[Security](https://worca.dev/security/)

**Autonomy, with limits.**

Guardrail policies that repo settings can't undo, a disposable container, and agents that never see a model key.

Security →
## From your laptop

to your team's server.

The same Worca everywhere. Start on your own machine and move to a shared server when the team joins, keeping your workflows and history.

**One npm install**

macOS, Linux or native Windows. `npm i -g @worca/app`

**A disposable Linux box**

Agents run inside it, and your home folder isn't mounted. `worca container up`

**One Worca, behind sign-in**

Any container host, behind the sign-in you already use. Worca records who started, paused and approved each run.

**Metrics and policy in git**

Spend, a delivery timeline and team caps, stored on branches in your own repo. No extra server.

## The rest of

the toolbox.

**Isolated worktrees** Each run gets its own git worktree and branch. Your checkout is never touched; parallel runs don't collide.

**Runs pause, they don't die** A 500, a rate limit or an expired login pauses the run with its worktree intact. Resume picks up where it stopped.

**Mock mode** The whole pipeline runs offline with a deterministic mock: no agents, no tokens. For demos, development and CI.

**Plugins & marketplaces** Task sources, agents, scripts, skills, workflows, models and chat channels. Each install shows what it adds and asks you first.

**Workspaces** Group related repos. One run fans out across all of them with a single cross-project review verdict.

**One-click PRs** A finished run shows its diff and opens a pull request, even from a fork, and reports back to the issue it came from.

**Nothing in your repo** Run state lives in one SQLite database in your home directory. Your checkout stays clean.

**Simple to Expert** New installs start with a short sidebar and a Getting started checklist. Switch to Advanced or Expert when you want everything.

## Ship work you

didn't babysit.

One npm install on macOS, Linux or Windows. Or a container, or a shared server for the team.