Pizza Bot solves the common problem of keeping AI agents productive while you focus on other work. Many teams struggle with scattered results, forgotten follow‑ups, and the need to constantly switch contexts to check on background tasks. Pizza Bot provides a single email‑style inbox that separates completed work, pending approvals, and full thread history, so you never lose track of what an agent has done or what still needs your input.
The tool works as a self‑hosted application you can run on macOS, Windows, or Linux, or access through a browser or terminal client. Its backend owns scheduling, so if your machine goes to sleep or you close the desktop app, missed cron intervals are collapsed into a single catch‑up run when the server restarts. This means you get reliable, continuous processing without worrying about duplicate work or lost progress.
Under the hood Pizza Bot uses DeepAgents and LangGraph for stateful execution. Checkpoints preserve thread state and approval pauses, allowing a task to survive a client disconnect as long as the backend stays online. If you quit the desktop app, the embedded server stops and any in‑flight step may be lost, but the thread itself remains recoverable when you reconnect to an always‑on backend.
You can connect the bot to a variety of model providers—Amazon Bedrock, Anthropic, Google Gemini, OpenAI, OpenRouter, or local Ollama models—by configuring them in the Settings > Providers menu. The agent includes a sandboxed JavaScript interpreter and scratch‑file operations, and it can delegate work to skill workers defined in SKILL.md files. Each skill declares its required tools and can be gated with interruptOn and allowedDecisions, letting you approve, edit, or reject specific actions before they run.
External tools are exposed through MCP servers, and existing Claude Code‑compatible configurations are supported. Skills are packaged with their MCP servers, making it easy to share and reuse specialized workers.
Key benefits for the audience:
– Organizes AI work in a clear inbox (All, Unread, Action) reducing context‑switching fatigue.
– Guarantees background tasks continue after you close the client, provided the backend stays up.
– Supports multiple LLM providers and local models, giving flexibility to choose cost, performance, or privacy preferences.
– Provides granular, tool‑level approval controls so you maintain oversight without micromanaging every step.
– Offers a deployable, Apache 2.0 licensed solution with desktop builds, browser/terminal clients, and a standalone server option.
Deploy Pizza Bot, configure your preferred model, define the skills you need, and let the inbox handle the rest—so your AI agents keep working while you stay focused on what matters.
#AI #Product #Automation #LLM #SelfHosted #Productivity