ClawPilotRouter — Copilot Model Router

Routes every coding request to the best model for the task. Simple completions get a fast model, complex refactors get a strong one. Runs as a local OpenAI-compatible proxy — point your editor at it and go.

How It Works

1. You start ClawPilotRouter (one command)
2. Point your copilot/editor at http://127.0.0.1:8402/v1 with model auto
3. ClawPilotRouter classifies each request across 14 weighted dimensions in <1ms
4. Routes to the best copilot model for the task (4 tiers)
5. Falls back through alternatives if a provider errors

Tier	What it handles	Default Model	Multiplier
SIMPLE	Inline completions, lookups, short answers	GPT-4o	0x (free)
MEDIUM	Code generation, refactoring, explanations	Claude Sonnet 4.6	1x
COMPLEX	Multi-file edits, architecture, debugging	Claude Opus 4.6	3x
REASONING	Algorithm design, proofs, complex debugging	GPT-5.4	1x

Quick Start

npm install -g clawpilotrouter

# Start the router — authenticates via GitHub on first run
clawpilotrouter

On first run, you'll see:

Visit: https://github.com/login/device
Enter code: XXXX-XXXX

Open the link, enter the code, and you're done. The token is saved to ~/.clawpilotrouter/github_token and refreshed automatically. No API keys to manage.

Then configure your editor (see VS Code Setup below).

That's it. The router uses Claude Sonnet 4 for standard coding tasks and Claude Opus 4 for complex ones. Simple completions go to fast, cheap models. Reasoning tasks go to models with large context windows.

VS Code Setup

ClawPilotRouter works with VS Code via the Continue extension, which supports custom OpenAI-compatible endpoints.

1. Install the Continue extension from the VS Code marketplace
2. Open ~/.continue/config.yaml and add:

name: My Config
version: 0.0.1
schema: v1

models:
  - name: ClawPilotRouter
    provider: openai
    model: auto
    apiBase: http://127.0.0.1:8402/v1
    apiKey: dummy

3. Start the router: npx clawpilotrouter
4. Select ClawPilotRouter from Continue's model dropdown in the sidebar

The apiKey value is a placeholder — the router handles authentication via your GitHub token.

To pin a specific model instead of smart routing, replace auto with any model alias (e.g. sonnet, opus).

Usage Stats

ClawPilotRouter logs every routed request locally. Check your stats via the /stats endpoint:

curl http://127.0.0.1:8402/stats | python3 -m json.tool

Example output from the CLI:

╔════════════════════════════════════════════════════════════╗
║              ClawPilotRouter — Copilot Stats                    ║
╠════════════════════════════════════════════════════════════╣
║  Period: last 7 days                                       ║
║  Requests routed: 312                                      ║
║  Actual cost: $4.18                                        ║
║  If all Opus: $47.62                                       ║
║  You saved:   $43.44 (91%)                                 ║
║  Avg cost:    $0.0134/req                                  ║
╠════════════════════════════════════════════════════════════╣
║  Task distribution:                                        ║
║    Quick      ████████████         52% (162 reqs)          ║
║    Standard   ██████               28% (87 reqs)           ║
║    Complex    ███                  13% (41 reqs)           ║
║    Reasoning  █                     7% (22 reqs)           ║
╠════════════════════════════════════════════════════════════╣
║  Models used:                                              ║
║    grok-code-fast-1                 148 reqs  $0.31        ║
║    claude-sonnet-4.6                 92 reqs  $1.84        ║
║    claude-opus-4.6                   41 reqs  $1.62        ║
║    gemini-3.1-pro                    22 reqs  $0.38        ║
║    gpt-4.1                            9 reqs  $0.12        ║
╚════════════════════════════════════════════════════════════╝

The "If all Opus" line shows what it would cost to send every request to Claude Opus 4 — the premium baseline. The difference is what smart routing saves you.

Agentic Auto-Detection

ClawPilotRouter detects multi-step coding tasks and routes to models optimized for autonomous execution:

"what does this function do"           → grok-code-fast (SIMPLE)
"refactor this module to use async"    → claude-sonnet-4.6 (MEDIUM)
"fix the bug and run the tests"        → claude-opus-4.6 (COMPLEX, agentic)

No config needed — agentic detection works automatically based on keyword signals like file operations, execution commands, and iterative patterns.

Session Pinning

Multi-turn conversations stay on the same model to prevent mid-task switching:

Turn 1: "Build a React component"  → claude-opus-4.6
Turn 2: "Add dark mode support"    → claude-opus-4.6 (pinned)
Turn 3: "Now add tests"            → claude-opus-4.6 (pinned)

Sessions persist for 1 hour of inactivity.

Model Aliases

Pin a specific model instead of using auto:

Alias	Model
sonnet	Claude Sonnet 4.6
opus	Claude Opus 4.6
haiku	Claude Haiku 4.5
gpt	GPT-4.1
gpt5	GPT-5.4
codex	GPT-5.3 Codex
flash	Gemini 3 Flash
gemini	Gemini 3.1 Pro
grok-code	Grok Code Fast

CLI Options

clawpilotrouter                  # Start (authenticates on first run)
clawpilotrouter --port 9000      # Custom port
clawpilotrouter --version        # Show version
clawpilotrouter --help           # Show help

Port can also be set via CLAWPILOTROUTER_PORT environment variable.

Authentication

On first run, ClawPilotRouter authenticates via GitHub OAuth device flow — you visit a URL and enter a code. The token is saved to ~/.clawpilotrouter/github_token and refreshed automatically.

For non-interactive environments, set GH_TOKEN or COPILOT_GITHUB_TOKEN as an environment variable.

License

MIT