Agent Backends
PocketPaw supports three agent backends. Each provides different capabilities and trade-offs.
Backend Comparison
| Feature | Claude Agent SDK | PocketPaw Native | Open Interpreter |
|---|---|---|---|
| Provider | Anthropic Claude | Anthropic + OI | Ollama/OpenAI/Anthropic |
| Built-in tools | Bash, Read, Write, Edit | Custom + OI execution | OI tools |
| Local models | No | No | Yes (Ollama) |
| Code execution | Claude tools | Open Interpreter | Open Interpreter |
| Streaming | Yes | Yes | Yes |
| MCP support | Native | Via tool registry | Via tool registry |
| Recommended for | Coding, complex tasks | Balanced workflows | Local/offline use |
Switching Backends
Set the backend via environment variable or config:
export POCKETCLAW_AGENT_BACKEND="claude_agent_sdk" # default# orexport POCKETCLAW_AGENT_BACKEND="pocketpaw_native"# orexport POCKETCLAW_AGENT_BACKEND="open_interpreter"Or change it in the web dashboard’s Settings panel.
Backend Details
Claude Agent SDK
The recommended backend with native Claude tools and MCP support.
PocketPaw Native
Custom orchestrator combining Anthropic SDK with Open Interpreter.
Open Interpreter
Standalone engine with Ollama support for fully local operation.
Was this page helpful?