Install the Manifest plugin for OpenClaw — an open-source LLM router that saves up to 70% on model costs.
Why Manifest
- Save costs — Routes each request to the cheapest model that can handle it. A 23-dimension scoring algorithm picks the right tier in under 2ms.
- Route LLM requests — Supports 300+ models across Anthropic, OpenAI, Google, DeepSeek, xAI, Mistral, and more. One endpoint, all providers.
- Add fallbacks — If a model fails or hits a rate limit, traffic shifts to an alternative automatically. No downtime, no code changes.
- Set limits — Configure spending caps and email alerts per hour, day, week, or month.
Workflow
1. Determine parameters
- Mode (optional, default
local):local— All data stays on your machine. No account needed. Dashboard athttp://127.0.0.1:2099.cloud— Dashboard at app.manifest.build. Requires an API key.
- Key (required for cloud mode): A
mnfst_*API key from app.manifest.build.
If the user doesn't specify a mode, default to local. If cloud mode and no key, ask for it.
2. Run the install script
# Local mode (default — no account needed):
bash skills/openclaw-model-router/scripts/install_manifest.sh
# Cloud mode:
bash skills/openclaw-model-router/scripts/install_manifest.sh --mode cloud --key mnfst_YOUR_KEY
Use --dry-run to preview changes without modifying anything.
3. Show status table
After installation, run the diagnostic table:
bash skills/manifest-status/scripts/manifest_status.sh
Output the table exactly as printed. No extra commentary.
4. Next steps
Tell the user:
- Dashboard is available at http://127.0.0.1:2099 (local) or app.manifest.build (cloud)
- The default model is now
manifest/auto— requests are routed based on complexity tier - Customize routing tiers and spending limits in the dashboard