pydantic/pydantic-ai-harness

Your agent's favorite harness, built on Pydantic AI

What it solves

Pydantic AI Harness provides a comprehensive library of "capabilities" to transform simple AI agents into sophisticated, long-running autonomous agents. It solves the problem of agents degrading over time or lacking the the necessary environment to perform complex tasks like codebase maintenance, deep web research, or cross-session memory management.

How it works

The library operates on a primitive called a "capability"—a self-contained unit of agent behavior that can be added to any Pydantic AI agent via the capabilities=[...] argument. These capabilities are modular and composable, allowing developers to either snap on individual blocks (like a file system or a shell) or use pre-built "harnesses" (like the Coder or Researcher stacks) which are themselves combinations of multiple capabilities.

Who it’s for

Developers building autonomous agents with Pydantic AI who need their agents to have persistent memory, secure execution environments, planning abilities, and advanced context management to handle complex, unattended work.

Highlights

  • Pre-built Agent Stacks: Includes complete configurations for coding agents (Coder) and research agents (Researcher).
  • Execution Environments: Provides secure file system access, allowlisted shell execution, and isolated cloud sandboxes via Modal.
  • Context Management: Features tools for tool-result clearing, sliding-window trimming, and tool output limits to prevent context window bloat.
  • Reasoning and Delegation: Supports structured task planning, sub-agent delegation, and dynamic workflows where models orchestrate other agents.
  • Knowledge and Memory: Implements persistent namespaced notebooks, BM25 search over conversation history, and on-demand loading of skill instructions.
  • Safety and Control: Includes guardrails for input/output validation, prompt injection defense, and spend limits for token/USD budgets.

Related

  • Project
    vstorm-co/pydantic-deepagents

    A Python framework and terminal AI assistant that provides a complete agent harness with a unique live run forking capability to test parallel execution paths.

  • 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
    K-Dense-AI/agentic-data-scientist

    An adaptive multi-agent framework that automates complex data science workflows by separating planning from execution and utilizing a self-correcting, iterative validation loop.

  • Project
    MervinPraison/PraisonAI

    PraisonAI is an open‑source framework for building autonomous, self‑improving AI agents. It implements a five‑layer stack (prompt, context, harness, loop, graph) plus a managed‑agents layer that lets you run agents locally or in remote sandboxes (Docker, e2b, Modal, etc.). The Python SDK (`praisonaiagents`) provides declarative agents, tool integration via the MCP protocol, loop‑control (budgets, doom‑loop detection), and graph orchestration (routing, parallel, repeat). A CLI, visual flow builder, dashboard, and JavaScript SDK round out the ecosystem. Supports 100+ LLM providers and is MIT‑licensed.

  • Project
    dapr/dapr-agents

    Dapr Agents is a Python SDK built on the Dapr runtime that lets developers create scalable, fault‑tolerant AI agents. It provides durable workflows, actor‑based execution, built‑in data source bindings, multi‑agent communication, and automatic tool discovery, all with Kubernetes‑native deployment and strong security.