Chen-zexi/open-ptc-agent

Executes Python code in isolated Daytona sandbox environments with automatic MCP tool conversion, enabling agents to process large datasets locally and return only final results—reducing token consumption by 85-98% compared to traditional tool calling. Integrates with LangChain DeepAgents framework, supports multiple LLM providers (Anthropic, OpenAI, etc.), and includes built-in middleware for subagent orchestration, multimodal vision analysis, and file operations. Features progressive tool discovery, custom MCP deployments, and auto-upload of generated artifacts to cloud storage (S3, R2, Alibaba OSS).

693 stars.