vstorm-co/pydantic-deepagents

Open-source, self-hosted Claude Code - a terminal AI assistant and the Python framework behind it. Tool-calling, sandboxed execution, multi-agent teams, skills, checkpoints, unlimited context - on Pydantic AI, any model.

What it solves

Pydantic Deep Agents provides a comprehensive "agent harness" that transforms a raw LLM into a fully functional autonomous agent. It eliminates the need for developers to manually build repetitive infrastructure like planning, memory, sandboxed execution, and tool-calling for every new AI agent project.

How it works

Built on Pydantic AI, the project offers two primary interfaces: a terminal-based AI assistant (TUI) and a Python framework. It integrates a suite of capabilities including a copy-on-write filesystem for isolated execution, a multi-agent swarm for parallel task delegation, and a persistent memory system. A standout feature is "Live Run Forking," which allows an agent to branch a single execution into multiple parallel paths to test different approaches, with an AI judge or test results determining the winning branch to merge back into the main run.

Who it’s for

  • Developers who want a self-hosted, open-source alternative to Claude Code that works with any LLM provider.
  • AI Engineers looking for a type-safe Python framework to build complex research or coding agents without wiring the plumbing from scratch.

Highlights

  • Live Run Forking: Branch executions into parallel paths with isolated filesystems and budgets, merged by an AI judge.
  • Multi-Agent Swarm: Ability to spawn sub-agents for parallel workstreams with a shared message bus.
  • Sandboxed Execution: Docker-based workspaces for safe tool and command execution.
  • Persistent Memory: Session-spanning memory via MEMORY.md and domain-specific skills via SKILL.md.
  • MCP Support: Integration with Model Context Protocol servers (e.g., GitHub, Figma).
  • Type-Safe Structured Output: Native Pydantic model responses for full IDE autocomplete and validation.

Related

  • Project
    canyonroad/agentsh

    A secure, policy-enforced execution gateway for AI agents that intercepts and governs file, network, process, and database activity at runtime.

  • Project
    pydantic/pydantic-ai-harness

    Pydantic AI Harness is a modular library of 50+ “capabilities” that extend a Pydantic AI agent with tools, memory, planning, safety, SaaS integrations, durable execution, and observability. You compose the needed pieces (e.g., the ready‑made Coder harness) and drop them into an Agent’s `capabilities` list, enabling autonomous agents that can edit code, browse the web, interact with services, remember past work, stay within cost limits, and survive restarts.

  • Project
    pydantic/pydantic-ai

    A typed Python AI SDK for building extensible agents with structured outputs, durable execution, and support for multiple model providers and interfaces.

  • Project
    sandbaseai/sandbase-harness

    SandBase Harness is a self‑hosted runtime that adds persistent sessions, sandboxed tool execution, credential vaults, audit trails and a web console to AI agents. It implements a Claude‑Managed‑Agents‑compatible `/v1` API, stores everything locally in SQLite, supports Docker/K8s sandboxes, works with many model providers, and ships a TypeScript SDK for easy client integration.

  • Project
    strands-agents/harness-sdk

    An open-source SDK for building and running AI agents in Python and TypeScript, providing a model-agnostic framework for managing agent loops, tools, and memory.