What it does
Outsourcerer routes work from your main AI coding session to other engines you already have, such as Claude, Codex, Gemini, Devin, OpenRouter models, or local models on Ollama or LM Studio. It checks what is installed, picks a lane using live benchmark scores and what each model costs you, and offers the cheaper path before spending anything. You choose how much control to keep: auto-pilot, you-drive or hybrid. Delegated jobs can carry your skills along, with MCP servers on request on the Claude lanes. It can convene several stronger models as an advisor panel to review a plan or a diff and proceed only on consensus, fan agents out in parallel across backends, and run loops where a budget model does the attempts and your own tests do the checking. A ledger called the Tab keeps real cash spend and subscription plan usage in separate columns, and a fleet view lists Claude Code sessions and delegated jobs with live status. The site describes the tool as a self-contained bash script with no server, no proxy and no telemetry that calls the CLIs you already have installed.
Capabilities
- Routes each job across harnesses and models, including Claude Code, Codex, Antigravity, Devin, OpenRouter and local models
- Picks a model from live benchmark scores weighed against cost, and shows its reason
- Advisor panels: several strong models review a plan or diff in parallel and it proceeds only on consensus
- Carries your skills to the delegate, and MCP servers on request on the Claude lanes
- The Tab tracks cash spend and plan-limit usage separately per lane
- Fleet view of running sessions and delegated jobs with live status
When to use it
- You pay for several AI coding tools and want routine work moved off your most expensive model
- You want a second opinion from several models on a plan or diff before building
- You want repo mapping, mechanical refactors or test loops run on cheaper or local models
- You want a running record of what delegated work costs
