Show HN: Pizza Bot – An inbox for AI agents that work in the background
summary
Pizza Bot is an open-source, local-first inbox designed for managing long-running AI agent tasks asynchronously. By utilizing a stateful runtime and a clear separation between the backend server and multiple frontends, it allows users to start tasks and return to them later without losing progress. The system prioritizes user control through human-in-the-loop approvals and explicit local file access permissions. It is designed to be extensible via Model Context Protocol servers, custom skills, and plugins.
key points
- The application employs an asynchronous workflow where completed tasks land in an Unread queue and pending decisions land in an Action queue, ensuring work continues even when the client is disconnected.
- It supports a wide range of model providers including Amazon Bedrock, OpenAI, and Google Gemini, allowing users to bring their own API keys and preferred LLM backends.
- Security is handled via a local-first approach where the api-server binds to the loopback address by default and requires explicit user grants for accessing local directories.
who should read it
Developers and AI enthusiasts who want to build or use a local-first environment for managing complex, multi-step AI agent workflows.
discussion warm
consensus
Commenters generally view the project as a promising implementation of an asynchronous agent communication system. There is a shared agreement that agents require their own communication and task management interfaces, with the email-style inbox being an intuitive UX for knowledge workers. The sentiment is supportive, though some users questioned the necessity of a custom inbox over existing tools like real email or ticket trackers.
top insight
The most valuable perspective is that Pizza Bot serves as the execution harness rather than just a task source. While tools like Linear provide the 'what' via MCP servers, Pizza Bot manages the 'how' by consolidating multiple agent runs into a single inbox. This prevents the fragmentation of managing numerous terminal tabs or separate ticket threads when multiple agents are running concurrently.
Generated by an LLM from the article and thread. Check the sources before relying on it.