AWS has released Pizza Bot, an open-source application designed to run AI agents on your own computer or server, with built-in support for local models through Ollama and five major cloud-based AI providers. The tool organizes long-running background tasks in an email-style inbox, letting users review completed work, approve actions, and manage AI-powered workflows without relying on cloud services.

What Is Pizza Bot and Why Does It Matter?

Pizza Bot started as an internal Amazon tool used by more than 2,000 employees for meeting preparation, email drafting, Slack summaries, and research tasks. AWS rebuilt it as an open-source project under the Apache 2.0 license, making it freely available to anyone who wants to run AI agents locally. The application separates tasks into three categories: All (complete thread history), Unread (completed work awaiting review), and Action (work paused for approval or user input).

The tool's architecture is built for persistence and control. It uses DeepAgents and LangGraph, two frameworks for managing stateful AI workflows, to preserve task progress even if your computer disconnects or closes. A Hono API server handles runtime execution and storage, while SQLite databases store cross-thread memory and application settings. This means background tasks can continue running on an always-on backend server, even after you close the desktop application.

Which AI Models Can Pizza Bot Run?

Pizza Bot supports a diverse range of AI model providers, giving users flexibility in how they deploy their agents. The application works with Amazon Bedrock, Anthropic's Claude, Google Gemini, OpenAI, OpenRouter, and Ollama for local models. This multi-provider approach means you can choose between running models entirely on your own hardware using Ollama, or connecting to cloud-based services depending on your privacy, cost, and performance preferences.

The local-model option through Ollama is particularly significant for organizations concerned about data privacy or those operating in environments with limited internet connectivity. Users configure their preferred provider under Settings before running tasks, and the application handles all communication with the selected service.

How to Set Up and Use Pizza Bot for Your Workflow

• Choose Your Deployment Model: Pizza Bot offers macOS, Windows, and Linux desktop builds, plus browser and terminal clients that connect to a local or standalone backend server, giving you flexibility in how you access and manage your agents.
• Configure Your AI Provider: Before running tasks, select your preferred model provider (Ollama for local models, or cloud services like OpenAI and Anthropic) in the Settings menu to ensure agents use your chosen AI service.
• Set Task Triggers: Tasks can start manually through the interface, on automatic schedules using cron syntax, or through webhooks, allowing you to automate workflows based on external events or time-based triggers.
• Organize and Approve Work: Review completed tasks in the Unread folder, approve or reject proposed actions, and organize threads into custom folders to keep your agent work organized and auditable.

What Makes Pizza Bot Different From Running Agents in the Cloud?

Pizza Bot emphasizes control and transparency over convenience. The application includes a sandboxed JavaScript interpreter that agents can use for computation without access to your network or file system, protecting your machine from malicious code. Skill authors can configure approval gates for specific tools, requiring human review before sensitive actions like publishing content or accessing external systems.

The tool also supports Model Context Protocol (MCP) servers, which expose external tools and integrations to agents. Each skill is defined in a SKILL.md file that specifies the agent's instructions and which tools it can access. Skills only become callable when their declared dependencies are available, preventing agents from attempting actions they cannot perform.

One critical limitation: if you use the embedded desktop server, closing the application stops active runs. However, checkpoints preserve the thread state, so work can resume when you reopen the app. For truly uninterrupted background execution, Pizza Bot requires an always-on backend server running separately from the desktop client.

Why Local AI Agents Matter for Security and Privacy

The release of Pizza Bot reflects a broader shift in how organizations think about AI deployment. Running agents locally through Ollama means your data never leaves your infrastructure, addressing privacy concerns that arise when using cloud-based AI services. This is particularly important for companies handling sensitive information, such as healthcare providers, financial institutions, or organizations subject to data residency regulations.

The application's approval workflow also addresses a key concern in AI safety: ensuring humans remain in control of consequential decisions. By requiring explicit approval for certain tools before agents execute them, Pizza Bot gives users a way to audit and override AI decisions in real time.

Pizza Bot is available now under the Apache 2.0 open-source license, with code and documentation available on GitHub. The application is deployable on Windows, macOS, and Linux, and supports both desktop and server deployments, making it accessible to individual developers and enterprise teams alike.