01
Choose for each workload
Use a frontier model for difficult planning, a fast provider for routine turns, or a local model when privacy and offline access matter most.
// 20 providers
Bring your own model. Or all of them.
PasClaw normalizes a broad provider catalog behind one agent interface, making it practical to mix frontier APIs with local and self-hosted models.
20 providersFallback chainsCloud and localOne config
02
Fail over automatically
Ordered fallback chains can route around rate limits and transient provider failures without changing the surrounding application.
03
Keep tools consistent
PasClaw translates provider-specific tool calling, streaming, and response formats into the same internal loop.
04
Avoid platform lock-in
Provider selection lives in configuration, so applications and tool implementations do not need to be rewritten when models change.
From the documentation
Technical details
Implementation notes drawn from the PasClaw repository documentation.
One catalog, three protocols
The catalog groups cloud, local, and self-hosted providers behind Anthropic, OpenAI, and Gemini protocol families.
- Bearer, header-based, and unauthenticated local endpoints are supported.
- Ordered fallbacks live in the fallbacks array in config.json.
- Adding a provider is designed as a catalog record rather than a new agent loop.
Caching and search
Provider-specific caching controls reduce repeated prompt cost, while web search is configured as its own provider layer.
- Anthropic cache_control and OpenAI prompt_cache_key are supported.
- Search options include DuckDuckGo, Brave, Tavily, SearXNG, Perplexity, and Gemini grounding.
- The current documentation lists 20 model-provider options, including xAI and LM Studio.
Source documentation
Common questions
Frequently asked questions
Keep exploring
See how the rest of PasClaw fits together.