Providers & Agents

ClawRun acts as a translation layer between agents and providers. Both interfaces are pluggable, so you're never locked into a single runtime or a single agent.

Agents

ClawRun's Agent interface is pluggable. Implementations self-register via the agent registry.

Agent	Status	Description
ZeroClaw	Supported	Fully autonomous AI agent with tools, memory, and cron support
OpenClaw	Coming soon
NanoClaw	Coming soon

Providers

ClawRun's SandboxProvider interface abstracts sandbox creation, snapshotting, and lifecycle management.

Provider	Status	Description
Vercel	Supported	Firecracker microVMs with sub-second cold start and filesystem snapshots
Cloudflare	Coming soon
Netlify	Coming soon
Fly.io	Coming soon

LLM providers

ZeroClaw supports 48 LLM providers out of the box. Select your provider and model during clawrun deploy.

Provider	ID
OpenRouter	`openrouter`
Anthropic	`anthropic`
OpenAI	`openai`
OpenAI Codex	`openai-codex`
Google Gemini	`gemini`
DeepSeek	`deepseek`
Mistral	`mistral`
xAI / Grok	`xai`
Perplexity	`perplexity`
Venice AI	`venice`

Fast inference

Provider	ID
Groq	`groq`
Fireworks AI	`fireworks`
Together AI	`together-ai`
Novita AI	`novita`
NVIDIA NIM	`nvidia`

Gateway / Proxy

Provider	ID
Vercel AI Gateway	`vercel`
Cloudflare AI Gateway	`cloudflare`
Amazon Bedrock	`bedrock`
Astrai	`astrai`

Specialized

Provider	ID
Cohere	`cohere`
Kimi Code	`kimi-code`
Qwen Code	`qwen-code`
Qwen (DashScope, China)	`qwen`
Qwen (International)	`qwen-intl`
Qwen (US)	`qwen-us`
Qwen Coding Plan	`qwen-coding-plan`
Moonshot (China)	`moonshot`
Moonshot (International)	`moonshot-intl`
GLM / Zhipu (International)	`glm`
GLM / Zhipu (China)	`glm-cn`
MiniMax (International)	`minimax`
MiniMax (China)	`minimax-cn`
Hunyuan / Tencent	`hunyuan`
Qianfan / Baidu	`qianfan`
Z.AI (Global)	`zai`
Z.AI (China)	`zai-cn`
Synthetic	`synthetic`
OpenCode Zen	`opencode`

Local / Private

Provider	ID
Ollama	`ollama`
LM Studio	`lmstudio`
llama.cpp	`llamacpp`
SGLang	`sglang`
vLLM	`vllm`
Osaurus	`osaurus`

ClawRun supports 15 messaging channels. Channels with wake-hook adapters can wake a sleeping sandbox when a message arrives. Other channels are configured inside the agent and work when the sandbox is running.

Wake-hook enabled

These channels can wake a sleeping sandbox when a message arrives:

Channel	ID
Telegram	`telegram`
Discord	`discord`
Slack	`slack`
WhatsApp	`whatsapp`
Lark / Feishu	`lark`
QQ	`qq`

Agent-side only

These channels are configured in the agent but do not have wake-hook adapters yet:

Channel	ID
iMessage	`imessage`
Matrix	`matrix`
Signal	`signal`
Linq	`linq`
IRC	`irc`
DingTalk	`dingtalk`
Nostr	`nostr`
Nextcloud Talk	`nextcloud_talk`
Webhook (generic)	`webhook`

Tools

Tools are installed inside the sandbox at startup. Select which tools to enable during clawrun deploy.

Tool	ID	Description
Agent Browser	`agent-browser`	Headless Chromium browser for web browsing and screenshots
GitHub CLI	`gh-cli`	GitHub CLI for managing repos, issues, and PRs