Any model. Every dollar accounted.
Run pipelines on Claude, on your GitHub Copilot subscription, on a model on your own machine or on free OpenRouter models. Pick per agent, per workflow or per run, and see what each step cost and what it saved.
Bring the models
you already pay for.
A model bridge is built in, so there's no proxy to install and no second service to run. Sign in or point Worca at a server, import the models, and pick them anywhere you pick a model.
Import asks the server which context window it really serves, so a prompt never outgrows the model.
Your models,
your routing.
Register any model id (a proxy, a fine-tune, another provider), say which effort levels it supports, and attach the routing it needs. No first-party Anthropic account is required.
The Models view: endpoint-routed entries are labelled on the card.
- ◎Model & effort per agentA cheap model for the planner, a strong one for the reviewer. Per agent, per workflow or per run.
- ⇢Every code path follows your endpointTitles are written by the run's own model, and built-ins can be hidden from every picker.
- ▶Test before you trustA Test button checks a catalog model answers before a run depends on it.
- $Prices where the endpoint has noneAn opt-in per-model cost override keeps estimates honest. Duplicate a model to derive a sibling.
- ⇪Share a catalog as a pluginModels, their pricing and the secrets they need, installed in one step.
Spend you can
see, and cap.
Cost estimates per run, per step and per agent, visible while the run is still going. A spend indicator sits in the sidebar, and limits pause work before a surprise shows up on the bill.
- ⏸Per-pipeline cost capA runaway run pauses itself and resumes only with an explicit override.
- ■Total budgetPauses everything and blocks new runs until the weekly or monthly window resets.
- ⚇Team capsA team policy can set caps for everyone. Going past one takes a reason, and the team sees it. Team policy →
- ▥StatisticsSpend, time worked and outcomes per day, week or month.
Settings › Runs: cost limits, schedules and notifications in one tab.
Is it worth
what it costs?
Every step of a run is credited with the hours a developer would have needed for what it produced. Statistics and the sidebar show Saved next to Spent: those hours at your hourly rate, minus what the run cost.
Statistics: Saved beside Spent, with the hours of work behind it.
One run, credited step by step
No model call is used to estimate it. A failed run that produced nothing counts against you.
Free requests,
counted.
The sidebar shows how many of today's free OpenRouter requests are left. When they run out, the run pauses at once and says when they come back, instead of retrying all night.
- Each call on a free model counts downYou're warned before a run starts if too few are left.
- At zero the run pauses, no retriesWith the reset time, in plain words.
- Resume after the reset, or pick a paid modelNothing is lost: the worktree and the work so far are kept.
Under this month's spend: today's free requests left.
Ship work you
didn't babysit.
One npm install on macOS, Linux or Windows. Or a container, or a shared server for the team.