Open Interpreter

The Open Interpreter backend is a standalone execution engine that supports multiple LLM providers, including Ollama for fully local operation.

Overview

Open Interpreter provides:

• Multi-provider support: Ollama, OpenAI, Anthropic
• Local models: Run with Ollama for completely local, offline operation
• Code execution: Built-in code interpreter for Python, JavaScript, and shell
• Streaming: Real-time response streaming

Configuration

export POCKETCLAW_AGENT_BACKEND="open_interpreter"

# For Ollama (local)
export POCKETCLAW_OI_MODEL="ollama/llama3.2"

# For OpenAI
export POCKETCLAW_OI_MODEL="gpt-4o"
export POCKETCLAW_OPENAI_API_KEY="sk-..."

# For Anthropic
export POCKETCLAW_OI_MODEL="claude-sonnet-4-5-20250929"
export POCKETCLAW_ANTHROPIC_API_KEY="sk-ant-..."

Local Operation with Ollama

For a completely self-contained setup with no external API calls:

curl -fsSL https://ollama.com/install.sh | sh

ollama pull llama3.2

export POCKETCLAW_AGENT_BACKEND="open_interpreter"
export POCKETCLAW_OI_MODEL="ollama/llama3.2"

pocketpaw

When to Use

Choose Open Interpreter when:

• You want fully local operation with no external API calls
• You’re running on hardware with a capable GPU for local model inference
• You want to use open-source models like Llama, Mistral, or Phi
• You need offline capability

Limitations

Compared to the Claude Agent SDK backend:

• No native MCP support (tools registered via tool registry)
• Tool calling depends on the model’s capabilities
• Smaller models may not reliably use complex tool chains
• No built-in safety hooks (relies on PocketPaw’s Guardian AI)

Installation

Open Interpreter is included in the core installation:

curl -fsSL https://pocketpaw.xyz/install.sh | sh

For Ollama, install it separately from ollama.com.