Outsourcerer
Bash-based orchestrator that routes coding tasks across Claude, Codex, Cursor, Devin and local models while tracking spend in one ledger
177+
GitHub Stars
6+
Engines Orchestrated
2026
Founded
Overview
Outsourcerer is an open-source bash script that coordinates multiple AI coding assistants, Claude Code, Codex, Cursor, Devin, and local Ollama/LM Studio models, so they work as one team instead of as separate disconnected tools. It evaluates each incoming task against live benchmark scores and routes "boring" work (grep, refactors, test runs) to cheaper or local models while reserving premium models for judgment calls. A built-in ledger ("the Tab") tracks real per-generation API spend alongside subscription quota usage, and the tool can carry an entire setup (skills, docs, scripts) across engines when delegating a task.
The Verdict
Who Should Use Outsourcerer?
Best For
- Developers juggling multiple AI coding subscriptions who want one cost view
- Teams running parallel agent fleets across different backends
- Anyone who wants cheap/local models to absorb mechanical work automatically
- Users who want bounded retry loops validated by their own test suite
- Command-line-comfortable developers willing to run a bash tool
Not Ideal For
- Commercial users (license requires separate commercial agreement)
- Teams wanting a polished GUI instead of a CLI/plugin workflow
- Anyone needing full visibility into non-Claude Code agent sessions (fleet heartbeat is limited)
- Developers who only use a single AI coding tool
- Windows-native workflows without a bash/WSL environment
What's Great
- Routes tasks to the cheapest capable model instead of always using the priciest one
- Single ledger tracks both pay-per-use API cash and subscription plan-limit burn
- Carries skills, docs, and context across engines on delegation
- Advisory panels require multi-model consensus before work proceeds
- Local model support (Ollama, LM Studio) for fully private, $0 lanes
- Hard credential blocks prevent .env/id_rsa files from leaving the machine
Watch Out For
- PolyForm Noncommercial license, commercial use needs a separate license from the author
- Fleet visibility currently covers only Claude Code and Outsourcerer-run jobs
- Bounded loops need measurable pass/fail criteria; no good for subjective quality checks
- Shell-script distribution rather than a packaged binary or app
- Free (":free") model lanes may train on your prompts unless scoped out
Pricing
Open Source
$0
Free for noncommercial use under PolyForm Noncommercial 1.0.0
Commercial License
Contact
Required for commercial use, contact the author directly
Pass-Through API Costs
Variable
You pay OpenRouter/provider rates for actual model usage it routes to
View all features & details
Key Features
- Benchmark-driven task-to-model routing
- Context portability across engines (skills, docs, scripts)
- Multi-model advisory panels requiring consensus
- Bounded iteration loops (build-test-retry) with test-verified stopping
- Auto-pilot, you-drive, and hybrid approval modes
- Consolidated fleet view of active AI sessions
Supported Engines
- Claude Code (native plugin)
- Codex (parity bridge)
- Cursor, Droid, Hermes (reverse-bridge)
- Devin (skill mirroring)
- Ollama, LM Studio (local, private)
Use Cases
- Offloading mechanical work like repo mapping and large refactors
- Convening expert review panels before building
- Running parallel agent teams across backends
- Bounded retry loops verified by real test suites
- Keeping premium models reserved for high-judgment decisions
Security
- Hard blocks on credential files leaving the machine during delegation
- Per-repo trust scoping for narrow exceptions
- Least-privilege defaults for free/training-flagged model lanes
How It Compares
| Feature | Outsourcerer | LiteLLM Proxy | OpenRouter | Portkey |
|---|---|---|---|---|
| Price Model | Free (noncommercial) | Free (OSS) | Pay-per-use | Freemium |
| Routes | Whole coding agents/engines | Raw LLM API calls | Raw LLM API calls | Raw LLM API calls |
| Cost Tracking | Unified ledger (cash + plan limits) | Spend tracking | Pay-per-use billing | Spend tracking |
| Local Models | Yes (Ollama, LM Studio) | Via config | No | Via config |
| Multi-Agent Panels | Yes, consensus-gated | No | No | No |
| Context Portability | Yes, across engines | No | No | No |
| Commercial Use | Requires license | Yes (MIT) | Yes | Yes |
| Distribution | Bash script/plugin | Self-hosted proxy | Hosted API | Self-hosted or managed |
User Reviews
Loading reviews...