Bring any LLM — and switch per agent
Run FastClaw agents on OpenAI, Anthropic, DeepSeek, Ollama, OpenRouter, Groq, Mistral or any OpenAI-compatible API, with per-agent overrides and retries.
Any OpenAI-compatible API
Presets for OpenAI, Anthropic, Ollama, OpenRouter, Groq, DeepSeek and Mistral — plus any endpoint that speaks the OpenAI API.
Per-agent overrides
Set a system default, then give any agent its own provider and model. Agent-scope entries shadow system ones by name.
Inherit and layer
Non-admin users see admin-shared providers as Inherited and can add private ones on top — the same inheritance the runtime uses.
Prompt cache aware
Raw assistant turns are preserved so provider prompt caches keep hitting. Usage reports cache reads and writes separately.
Survives flaky APIs
Timeouts, dropped connections, 429s and 5xx are retried up to 6 times with 2s to 30s backoff instead of ending the turn.
Images in, images out
Send images as OpenAI image_url or Anthropic base64 parts; pictures in replies render in the console and IM chats.
One runtime, every provider
Configure providers once in the dashboard under Models, or from the CLI:
fastclaw agents config alpha set provider.openai.apiKeyEnv OPENAI_API_KEY
fastclaw agents config alpha set provider.openrouter.apiBase https://openrouter.ai/api/v1
Presets fill in the base URL and defaults for OpenAI, Anthropic, DeepSeek, Ollama, OpenRouter, Groq and Mistral. --api-key-env reads the key from an environment variable so it never sits in shell history.
Defaults with overrides
The system default model applies to every agent that doesn't say otherwise. An agent's own model setting overrides it, and agent-scope provider entries shadow system entries with the same name — so one agent can use a different key or base URL without affecting the rest. This is how SaaS builders offer tiers and how developers mix local and hosted models.
Built for long agent loops
Agent turns can run many tool calls, so FastClaw is tolerant of model hiccups. Requests retry with backoff, a reply cut off at the output limit is retried with a note to work in smaller steps, and DeepSeek thinking mode passes reasoning_content back correctly. Every turn reports real token usage — tool loops included — through the runtime API.
Works with
All Integrations
OpenAI
Run FastClaw agents on OpenAI models — set a deployment-wide default, override per agent, and expose your agents through an OpenAI-compatible API.
Anthropic
Run FastClaw agents on Claude through Anthropic's native Messages API — with prompt caching, extended thinking and image input preserved.
DeepSeek
Run FastClaw agents on DeepSeek — a built-in preset, thinking-mode support and per-agent model overrides for low-cost, capable agents.
Ollama
Point FastClaw at Ollama to run AI agents on local, open-weight models — no API key, no data leaving your machine.
OpenRouter
Connect FastClaw to OpenRouter and give every agent access to models from many vendors with a single API key.
Frequently asked questions
More Features
All FeaturesAgents
Create and run many AI agents from one FastClaw install — each with its own SOUL.md personality, memory, skills, model and sharing settings.
Channels
Bind any FastClaw agent to Telegram, Discord, Slack, Feishu, WeCom, WeChat, WhatsApp, LINE or iMessage — with per-chat session isolation and owner pairing.
Tools & Sandbox
Built-in exec, file and web tools, run in an E2B cloud sandbox or a Docker container, with automatic skill and workspace hydration and file sync.
Try it on your own agents
Start free in FastClaw Cloud, or self-host the single binary in a minute.