Learn

Introduction

Mobius is a place to try agents without turning every experiment into infrastructure. Start with a message or a small loop, inspect what happened, and adjust until it fits the job.

If you have been running cron + claude -p, a Codex script, or a small MCP service that is starting to feel like infrastructure, Mobius is the hosted runtime for that shape. If you have been building Slack bots, Telegram bots, or embedded chat surfaces, Mobius gives those conversations durable sessions too.

Mobius has two first-class product shapes:

ShapeUse it whenDurable record
Agent messagingPeople should keep talking to an agent from Slack, Telegram, or your API.Agent session
LoopsThe agent should follow a repeatable process with triggers and steps.Run

Most teams start with a familiar use case: a morning brief, pull request review, issue triage pass, support channel, or in-product assistant. Mobius lets that story stay small at first, then grow into messaging bindings, schedules, provider actions, workers, and human review when the use case earns it.

That can be internal business operations or part of your product backend. The same project model supports both.

For internal business operations, pull request review is the example to keep in mind. The agent drafts a review, a person keeps ownership, and Mobius records the run before anything lands back in GitHub.

How you'll use it

Three surfaces share the same backend contracts:

SurfaceUse it for
The appTry agents, tune loops, inspect runs, and review sessions.
The mobius CLIOperate from a terminal or CI process.
The HTTP APIConnect Mobius to your own services.

Start in the app if you are new. Use the CLI when you need repeatable operator commands. Use the API when Mobius is part of another system.

Start here

  • Quickstart creates Scout, builds morning-brief, and starts a manual run.
  • Review every pull request is the first full internal operations recipe to try after the quickstart.
  • How Mobius works connects the core concepts and the two ways to use an agent.
  • Agent messaging explains Slack, Telegram, and API conversations.
  • Loops explains when useful agent work should become a repeatable process.
  • Agent sessions explains how one conversation keeps useful context over time.

Where things live

TrackCovers
LearnThe product model and the core runtime concepts.
GuidesEnd-to-end recipes and operator tasks.
APIHTTP authentication, agent invocation, event streams, and the generated reference.
CLIInstall, auth, and terminal operation notes.

If you remember one pattern, remember this: start small, try it once, change one thing. The same project can hold quick experiments, production loops, and channel agents.