AWS Launches Pizza Bot Open Source AI Agent Inbox

Amazon Web Services (AWS) introduced Pizza Bot, a self-hosted application designed to manage background AI agents and their asynchronous tasks. This open-source project organizes completed results and pending decisions into an email-style inbox, allowing users to continue working while AI agents handle tasks in the background. Earlier internal versions of Pizza Bot supported over 2,000 Amazon employees, assisting with functions such as meeting preparation, email drafting, summarizing Slack conversations, logging customer relationship management (CRM) data, and conducting research. The publicly released application has been rebuilt as an open-source project, licensed under the Apache 2.0 license, and is available for deployment.

Pizza Bot provides desktop builds for macOS, Windows, and Linux, alongside browser and terminal clients. These clients can connect to a local or standalone backend server. The application categorizes tasks into distinct sections: 'All' displays the complete thread history; 'Unread' shows completed work that requires user review; and 'Action' holds tasks that are paused awaiting approval or a response. Users can further organize these threads into custom folders and monitor the status of delegated workers within the 'Activity' panel. Task initiation can be manual, scheduled via cron jobs, or triggered by webhooks, with the server managing the scheduling process. In the event of downtime, missed cron intervals are handled by a single catch-up run, preventing the replaying of every missed interval, and trigger occurrences are durably recorded.

The runtime architecture of Pizza Bot leverages DeepAgents and LangGraph for stateful execution. A Hono API server manages runtime execution and data storage. The Electron and browser clients utilize a shared React interface, with all clients communicating with the server via HTTP and server-sent events. LangGraph checkpoints are employed to preserve thread state and approval pauses. Separate SQLite databases store cross-thread memory and application metadata. Reconnecting clients can replay buffered events. While closing a thread or disconnecting a client does not halt a running server, quitting the desktop application terminates its embedded server and any active runs. Checkpoints ensure thread preservation, but the step in progress at the time of quitting may be lost. For continuous operation after the desktop application exits, an always-on backend server is necessary.

Pizza Bot supports a range of AI models and services, including Amazon Bedrock, Anthropic, Google Gemini, and OpenAI. It also incorporates skills and tools that enable agents to interact with external services. The application's design prioritizes user control and visibility over background AI operations, aiming to streamline workflows by offloading routine or time-consuming tasks to AI agents while keeping users informed and in control of critical decisions.