pydantic-deepagents

pydantic-deep is a deep-agent framework built on pydantic-ai, with planning, filesystem and subagent capabilities. Its features, as listed in the README, are multiple backends (in-memory StateBackend, LocalBackend, DockerSandbox and CompositeBackend), toolsets for todos, console, subagents and skills, file uploads, a skills system with markdown prompts, structured output through Pydantic models, automatic conversation summarization, human-in-the-loop confirmation and streaming.

The project is modular: backends come from the separate `pydantic-ai-backend` package and the todo toolset from `pydantic-ai-todo`, both usable with any pydantic-ai agent. Install with `pip install pydantic-deep` (add the `[sandbox]` extra for Docker sandboxes). The quick start creates an agent with `create_deep_agent()`, builds dependencies with `create_default_deps(backend)` and runs `agent.run(...)`; further examples cover `output_type` for structured results, `run_with_files()` for uploads and a summarization processor that triggers at a token threshold such as 100k and keeps the last 20 messages.