OpenRouter Model Sync

Syncs models from the OpenClaw app leaderboard on OpenRouter into this openclaw installation.

Workflow

Step 1: Extract Model IDs via Browser

The app page is a React SPA -- must use browser tool, not web_fetch.

1. Open https://openrouter.ai/apps?url=https%3A%2F%2Fopenclaw.ai%2F in browser (profile: openclaw)
2. Take a snapshot, find and click "Show more" to expand full leaderboard
3. Take another snapshot after expansion
4. Extract model IDs from link hrefs -- they follow the pattern /<provider>/<model-name> (e.g. /moonshotai/kimi-k2.5)
5. Filter out non-model links (navigation links to /docs, /chat, /rankings, /pricing, /enterprise, /about, etc.)
6. Collect the model IDs (without leading slash)

Step 2: Run Sync Script

python3 scripts/sync-openrouter-models.py --models "model/id1,model/id2,..."

Options:

• --dry-run — preview changes without writing files
• --json — machine-readable JSON output on stdout
• stdin also accepted (one model ID per line)

The script:

1. Verifies each ID against the OpenRouter /api/v1/models catalog (rejects unverified)
2. Converts verified models to openclaw format (context window, pricing, reasoning, modalities)
3. Creates timestamped backups before writing any config file
4. Adds missing models to ~/.openclaw/agents/<agent>/agent/models.json and ~/.openclaw/openclaw.json
5. Generates aliases (from references/aliases.json or auto-derived)

Step 3: Restart Gateway

openclaw gateway restart

Environment Variables

Variable	Purpose	Default
OPENCLAW_DIR	Override openclaw directory	~/.openclaw
OPENCLAW_AGENT_DIR	Override agent directory	auto-detected
OPENROUTER_API_KEY	API key (falls back to config)	from config

Maintaining Aliases

Edit references/aliases.json to add or update short aliases for model IDs. The script loads this file at runtime. If missing, built-in defaults are used.

What Gets Updated

• ~/.openclaw/agents/main/agent/models.json — openrouter provider models
• ~/.openclaw/openclaw.json — openrouter provider models + aliases
• Backups created as <file>.bak.<timestamp> before each write

Limitations

• Additive only (does not remove delisted models)
• Reasoning detection uses heuristics (architecture field + known model families)
• Requires browser tool for Step 1 (app page is JS-rendered)