pizza-bot-app/pizza-bot
A local-first inbox for long-running AI agents, built with DeepAgents and LangGraph.
What it solves
Pizza Bot provides an asynchronous "inbox" for long-running AI agent tasks. It solves the problem of having to stay connected or wait in real-time for an AI to finish a complex job by allowing users to start tasks, disconnect, and return later to find completed work in an "Unread" queue or pending approvals in an "Action" queue.
How it works
It uses a stateful runtime based on DeepAgents/LangGraph to manage agent workflows. The system consists of an api-server that handles the logic and model communication, which can be accessed via an Electron desktop app, a web app, or a terminal CLI. Users can configure various model providers (such as Amazon Bedrock, OpenAI, and Anthropic) and extend the bot's capabilities using MCP servers, custom skills, and plugins.
Who it’s for
It is designed for users who need to delegate complex, multi-step AI tasks to specialized agents and want to manage those tasks asynchronously without being tethered to a chat interface.
Highlights
- Asynchronous Workflow: Work continues even if the user disconnects; completed tasks land in Unread and approval requests land in Action.
- Human-in-the-Loop: Built-in support for approval gates to control consequential actions.
- Extensible Architecture: Supports MCP servers and custom "skills" that act as tool-scoped subagents.
- Model Agnostic: Compatible with multiple providers including Bedrock, Anthropic, Google Gemini, OpenAI, OpenRouter, and Ollama.
- Local-First Security: Application state is stored locally, and access to local files must be explicitly granted.
Related
- Project
- Project
- Project
- Project