# Pydantic Deep Agents — Python Deep Agent Harness

> Batteries-included Python agent harness with tool-calling, memory, hooks, and a terminal assistant CLI; verified 788★, pushed 2026-05-14.

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

```bash
pip install "pydantic-deep[cli]"
pydantic-deep --help
# Optional one-line installer (README):
curl -fsSL https://raw.githubusercontent.com/vstorm-co/pydantic-deep/main/install.sh | bash
```

## Intro

Batteries-included Python agent harness with tool-calling, memory, hooks, and a terminal assistant CLI; verified 788★, pushed 2026-05-14.

**Best for:** Python teams building tool-using agents who want a Claude Code–style harness

**Works with:** Python 3.10+; works with multiple providers/models via Pydantic AI integrations (per README)

**Setup time:** 12-25 minutes

### Key facts (verified)

- GitHub: 788 stars · 88 forks · pushed 2026-05-14.
- License: MIT · owner avatar + repo URL verified via GitHub API.
- README-backed entrypoint: `pip install "pydantic-deep[cli]" && pydantic-deep --help`.

## Main

- Start with the CLI to validate your environment, then switch to the framework API once you’ve chosen your tool set and provider config.

- Use lifecycle hooks (PRE/POST_TOOL_USE) to add safety gates, logging, and policy checks around shell/file tools.

- Keep memory explicit: treat `MEMORY.md` (and similar context files) as audited inputs, not a dumping ground for secrets.

- Prefer “capabilities” over ad-hoc glue code: README frames tools/memory/sandbox as composable capabilities you can enable per agent.

### Source-backed notes

- README describes a batteries-included agent harness (tools, memory, sandboxing, unlimited context) plus a terminal assistant CLI.
- README documents a one-line installer (`install.sh`) and pip-based install for Python environments.
- README mentions Claude Code-style lifecycle hooks (PRE/POST_TOOL_USE) as a first-class feature.

### FAQ

- **Is it just a library?**: No — README includes both a CLI terminal assistant and a framework API for building production agents.
- **How do I install?**: Use `pip install "pydantic-deep[cli]"` or the one-line `install.sh` method described in README.
- **What’s the safest way to extend it?**: Add policy checks in lifecycle hooks and keep tool permissions scoped per agent.

## Source & Thanks

> Source: https://github.com/vstorm-co/pydantic-deepagents
> License: MIT
> GitHub stars: 788 · forks: 88

---

<!-- ZH -->

## Quick Use

```bash
pip install "pydantic-deep[cli]"
pydantic-deep --help
# Optional one-line installer (README):
curl -fsSL https://raw.githubusercontent.com/vstorm-co/pydantic-deep/main/install.sh | bash
```

## Intro

Pydantic Deep Agents 是一套 Python 深度 agent harness：工具调用、持久化记忆、生命周期 hooks 与终端助手 CLI；已验证 788★，更新于 2026-05-14。

**Best for:** 想用 Python 搭建类似 Claude Code 的可执行 agent harness 的团队

**Works with:** Python 3.10+；README 表示可接入多种 provider/model（基于 Pydantic AI）

**Setup time:** 12-25 minutes

### Key facts (verified)

- GitHub：788 stars · 88 forks；最近更新 2026-05-14。
- 许可证：MIT；作者头像与仓库链接均已通过 GitHub API 复核。
- README 中可对照的入口命令：`pip install "pydantic-deep[cli]" && pydantic-deep --help`。

## Main

- 先用 CLI 验证环境与配置，再进入框架 API 层，把工具集与 provider 配置固化下来。

- 利用 PRE/POST_TOOL_USE 生命周期 hooks 在 shell/file 等工具前后加安全门、审计日志与策略校验。

- 显式管理记忆与上下文：把 `MEMORY.md` 等文件当作需要审阅的输入，避免把密钥/隐私写进去。

- 优先用“capabilities”组合：README 将工具/记忆/sandbox 抽象为可启用能力，便于按 agent 粒度配置。

### Source-backed notes

- README 将其描述为“深度 agent harness”（工具、记忆、沙箱、无限上下文），并提供终端助手 CLI。
- README 给出 `install.sh` 一行安装与 pip 安装方式，适配常见 Python 环境。
- README 提到 Claude Code 风格的 PRE/POST_TOOL_USE 生命周期 hooks 作为核心能力。

### FAQ

- **它只是库吗？**：不是；README 同时提供终端助手 CLI 与用于生产 agent 的框架 API。
- **怎么安装？**：按 README 使用 `pip install "pydantic-deep[cli]"`，或用 `install.sh` 一行安装方式。
- **怎么更安全地扩展？**：在生命周期 hooks 里增加策略校验，并把工具权限按 agent 进行收敛。

## Source & Thanks

> Source: https://github.com/vstorm-co/pydantic-deepagents
> License: MIT
> GitHub stars: 788 · forks: 88


---
Source: https://tokrepo.com/en/workflows/pydantic-deep-agents-python-deep-agent-harness
Author: Agent Toolkit