AWS Releases Pizza Bot, an Open-Source Inbox for Autonomous AI Agents

More than two thousand Amazon employees tested early iterations of Pizza Bot before the tool transitioned into a separate community-driven endeavor.

AWS has unveiled Pizza Bot, a fresh open-source tool designed to furnish developers with an email-style inbox for overseeing AI agents operating in the background. The application tackles a fundamental challenge: traditional chat interfaces prove inadequate for agents whose operations persist after users step away from their desks.

Pizza Bot adopts familiar email conventions to solve this problem. Finished tasks surface as unread messages, while items requiring human judgment appear as actionable items. An Activity panel additionally reveals assignments delegated to specialized agents, displaying their tool usage and current status.

Scheduled agents run autonomously in the background and surface updates directly into your inbox for review and triage.

Joseph Dolivo, principal technologist at AWS Startups

In a LinkedIn announcement on Thursday, Joseph Dolivo, principal technologist at AWS Startups and co-creator of the project, emphasized that Pizza Bot removes the responsibility for monitoring agent activity from the user's shoulders. "Instead of you having to initiate every conversation or wait on a prompt, scheduled agents run autonomously in the background and surface updates directly into your inbox for review and triage," he stated.

The official launch blog post reinforces this philosophy by highlighting that user inattention represents a fundamental design principle. The creators note that "The interface assumes you are not watching." They elaborate: "Nothing else we've seen starts there, and that one assumption is what buys you pauses that outlast the session that created them, notifications worth acting on, and scheduled work that produces threads instead of logs."

A community project

Though originating within Amazon, Pizza Bot now operates as an independent community initiative rather than an official AWS product. The project maintains its own GitHub organization, distinct from Amazon's infrastructure, and operates without AWS backing or service-level guarantees—users manage their own deployments entirely.

Pizza Bot functions as a desktop application compatible with macOS, Windows, and Linux, alongside browser and terminal interfaces. The application launches a local Pizza Bot server by default on the user's machine, with developers selecting their preferred underlying model from options including Anthropic, Amazon Bedrock, Google Gemini, OpenAI, OpenRouter, or a self-hosted model via Ollama.

The tool supports expansion through MCP servers and Agent Skills. The included browser-automation skill, for instance, leverages Playwright MCP to browse and manipulate web pages.

The server can operate on a continuously running machine or within a container environment, enabling scheduled agents to function continuously even when the user's laptop is powered down. Threads remain accessible from different devices when the user reconnects later.

Under the hood: LangGraph and ambient agents

Pizza Bot's agent execution layer relies on DeepAgents, LangChain's open-source framework for extended agent operations, which itself operates atop LangGraph, the runtime managing stateful agent workflows. Persistence emerges as the critical element: LangGraph preserves an agent's execution state throughout its operation, permitting runs to pause for approval, withstand client disconnections, and resume later without restarting from the beginning. Pizza Bot maintains these checkpoints, along with conversation threads and other data, on the local machine using SQLite and standard file storage.

AWS maintains its own open-source agents framework called Strands Agents, available since May 2025. Documentation indicates that AWS positions Strands as a "lighter-weight alternative to LangGraph for agents that don't need explicit graph control flow."

When asked via LinkedIn by The New Stack whether Strands could have served Pizza Bot's needs, Dolivo acknowledged the possibility, particularly given Strands' current TypeScript and workflow support. However, the team selected LangGraph "due to the maturity of the tooling and breadth of the exosystem," he explained. "It's also more familiar to many developers, and we wanted to reduce friction for community adoption since we knew we'd be open-sourcing it," he added.

Pizza Bot shares considerable similarity with a concept LangChain introduced in January 2025, when CEO Harrison Chase coined the phrase "ambient agents" to describe agents capable of responding to events, executing concurrently, and involving humans only when necessary. LangChain's prototype implementation featured an email assistant constructed on LangGraph. The company also created an "Agent Inbox": a dedicated interface modeled after email and customer-support platforms for managing ongoing interactions between users and background agents.

From 'JoeBot' to Pizza Bot

Pizza Bot's origins trace to April 2025, when Dolivo launched what he characterizes as a "side-of-desk passion project" called "JoeBot" aimed at streamlining routine CRM data entry. He subsequently partnered with colleague Igor Fil to transform that initial script into Pizza Bot, an MCP server capable of running parameterized, deterministic "recipes" throughout Amazon's internal systems.

The project's popularity expanded beyond its engineering originators into less-technical departments. As Pizza Bot matured, Dolivo notes it attracted more than 30 contributors and surpassed 2,000 users throughout Amazon, necessitating a user-friendly interface that non-technical staff could access.

"An MCP server requires an MCP client, and expecting non-technical users to work out of an IDE or terminal was never going to cut it," Dolivo remarks. "We had to meet people where they actually work and own the experience end to end."

The outcome was this week's Pizza Bot release: a desktop application structured around an inbox paradigm rather than command-line interaction.

Source: The New Stack