Pizza Bot checkpoints agent state to SQLite so a paused approval outlives the session

The inbox is what you see. The mechanism underneath is the interrupt: the agent graph stops at a declared point, LangGraph writes the state to a checkpoint, and the run resumes from that record later instead of starting again from scratch ^[14]. Pizza Bot leans hard on that persistence and those interrupts for its pauses and resumable threads ^[19]. The checkpoints, the threads and the rest of its application data sit locally, in SQLite and ordinary files ^[15].

Joseph Dolivo is a principal technologist at AWS Startups and a co-creator. He wrote in a LinkedIn post on Thursday: "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" ^[4]. The launch blog post states the design brief more bluntly. "The interface assumes you are not watching," the creators wrote ^[5].

The desktop app runs on macOS, Windows and Linux, and by default it starts a local Pizza Bot server on the machine ^[8]. A laptop nobody is watching is often also a laptop that is asleep. The server can instead run on an always-on host or in a container. That is what keeps scheduled agents working while the laptop is closed, and it lets their threads be picked up later from another device ^[12].

The rest is bring-your-own. Developers choose the model behind the app, and the project names six backends: Anthropic, Amazon Bedrock, Google Gemini, OpenAI, OpenRouter, and a local model via Ollama ^[9][10]. Extension runs through MCP servers and Agent Skills, and the bundled browser-automation skill uses Playwright MCP to navigate and interact with websites ^[11]. There is no AWS support and no service-level agreement, because the project now lives in its own GitHub organization, separate from Amazon, and is entirely self-hosted ^[7].

AWS has had its own open source agents SDK, Strands Agents, out since May 2025 ^[16], while Pizza Bot's runtime is DeepAgents, LangChain's harness for long-running agent tasks, running on LangGraph ^[13]. The New Stack reports that evidence including text in an AWS sample repository suggests AWS considers Strands a "lighter-weight alternative to LangGraph for agents that don't need explicit graph control flow" ^[17]. The same piece allows that a community project may simply never have been expected to use Strands ^[18].

The novelty claim is the weak part. "Nothing else we've seen starts there," the creators wrote, "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" ^[6]. In January 2025, LangChain CEO Harrison Chase introduced the term "ambient agents" for agents that could respond to events, work concurrently and involve a human only when needed. LangChain's reference implementation was an email assistant built on LangGraph ^[20]. LangChain also developed an Agent Inbox, a standalone interface inspired by email and customer-support software for keeping track of open interactions between people and background agents ^[21].