Applications¶
pydantic-deep is a framework — but the best way to see what it can do is to look at real apps built on it.
This section walks through four reference applications. Each one is a complete, working program that you can read, run, and copy from. They're not toy demos: they're the same create_deep_agent() you met in the tutorial, dressed in a real interface and pointed at a real job.
Learn by reading
Stuck on how to wire a feature into your own app? Open the reference app that uses it. The CLI shows streaming and sessions; DeepResearch shows subagents and plan mode; Harbor shows headless evaluation. Steal freely.
CLI — the terminal assistant¶
A Textual TUI that turns pydantic-deep into a Claude-Code-style coding assistant, right in your terminal. You get a live chat view, streaming output, tool-call diffs, slash commands, sessions you can fork and resume, and MCP support — all on top of the framework's defaults.
It's the fastest way to use a deep agent without writing any glue code.
DeepResearch¶
An autonomous research web app. You ask a question; it plans, searches the web, runs code, spawns subagents to chase down sub-questions in parallel, and assembles a cited answer — sketching its reasoning onto an Excalidraw canvas as it goes.
It's the showcase for the heavier features: subagents, plan mode, web search, and code execution working together in one flow.
ACP adapter¶
Run a pydantic-deep agent inside your editor. The ACP adapter speaks the Agent Client Protocol, so editors like Zed can drive your agent as a native coding assistant — same tools, same backend, no terminal required.
It's a thin bridge: point it at an agent, and your editor does the rest.
Harbor¶
A Terminal-Bench evaluation adapter. Harbor wraps a deep agent in the harness Terminal-Bench expects, so you can measure how your agent actually performs on real terminal tasks — headless, reproducible, scored.
It's how you turn "feels good" into a number.
Recap¶
- The reference apps are real programs built on the same framework you've been learning — proof of what you can ship.
- CLI is a terminal assistant; DeepResearch is an autonomous research web app; ACP runs agents inside your editor; Harbor benchmarks them.
- Read them when you want a working pattern for a feature, not just an API.
Pick one and dive in: