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.
- Open
https://openrouter.ai/apps?url=https%3A%2F%2Fopenclaw.ai%2Fin browser (profile: openclaw) - Take a snapshot, find and click "Show more" to expand full leaderboard
- Take another snapshot after expansion
- Extract model IDs from link hrefs -- they follow the pattern
/<provider>/<model-name>(e.g./moonshotai/kimi-k2.5) - Filter out non-model links (navigation links to
/docs,/chat,/rankings,/pricing,/enterprise,/about, etc.) - 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:
- Verifies each ID against the OpenRouter
/api/v1/modelscatalog (rejects unverified) - Converts verified models to openclaw format (context window, pricing, reasoning, modalities)
- Creates timestamped backups before writing any config file
- Adds missing models to
~/.openclaw/agents/<agent>/agent/models.jsonand~/.openclaw/openclaw.json - Generates aliases (from
references/aliases.jsonor 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)