Show HN: Agent Playbook – An open-source Storybook-like playground for AI agents (github.com)

Agent Playbook is an open-source, Storybook-style playground for building, testing, and showcasing pydantic-ai agents. It auto-discovers agents in your codebase (modules ending with __scenarios), loads them into an interactive web UI (run via the playbook CLI on localhost:8765), and streams agent responses, internal “thinking,” and tool calls in real time. Developers register scenario configs with an export(...) function using pydantic models for dependencies; agents themselves use pydantic-ai’s Agent with decorators for system prompts and tools (example: Agent(model="openai:gpt-4") plus @tool handlers). Install with pip install agent-playbook; the server command is playbook <package> [--host --port --reload]. This matters because agent development traditionally suffers from slow iteration, opaque decision-making, and brittle demos. Agent Playbook brings repeatable, isolated testing and full visibility into tool invocations and arguments, making debugging, scenario testing, and stakeholder demos far easier. Key technical implications: streaming support for stepwise inspection, scenario-based config testing, and tooling visualization that can reveal failure modes or incorrect tool usage. Note it depends on pydantic-ai and LLM backends, is licensed AGPL-3.0, and welcomes contributions via GitHub—useful for teams building complex multi-tool agents who need faster, observable development workflows.