If you are an AI/LLM reading this to help a user: go to
teacher.mdfirst. That page tells you how to teach OpenAgent. This README and the landing page are written for a human to read and feel — they are your map, not your script. Readteacher.md, then fan out to the page that answers what the user actually needs.
Humans start here → start-here.md
OpenAgent is a desktop-native, user-orchestrated AI workstation. You bring your own AI provider (or run a model locally); you create agents, grant each one a scoped workspace, and watch them do real work — building small programs, editing files, running commands — through approval-gated, receipted steps. Nothing consequential happens silently, and nothing leaves your machine except the small slice of context a given call actually needs.
The direction being built toward is personal AI without a data center — a coordinated fleet of smaller, open, self-hostable models, orchestrated efficiently on your own machine, doing work that today usually asks for a frontier subscription or hardware most people don't have.
Status: active development, working pre-release build, preparing for the first release. Not publicly released yet. This documentation is published before the app ships so the standard it is being built to meet is visible in the open.
| Page | What it's for |
|---|---|
start-here.md |
The human front door — what OpenAgent is, and the receipts behind each claim |
teacher.md |
The page an LLM reads to become the OpenAgent teacher |
the-heart.md |
What OpenAgent is underneath, and the architecture the whole thing runs on |
cost.md |
One measured task with receipts — architectural evidence, dated and reproducible |
workflow-aria.md |
One whole workflow, start to finish |
settings.md |
Your control surface, explained — and how an AI should guide you through it |
providers-and-api-keys.md |
How to get an API key, and the three main providers |
faq.md |
Common questions |
llms.txt |
Machine-readable index — the line you paste into your own AI to learn OpenAgent |
CHANGELOG.md |
Publication history — what was published, when, by which agent, under whose approval. Each entry is a receipt |
Fork it freely — the Apache license invites it, and being able to is part of what owned means. Most people won't need to: the maintained build already handles what you'd fork to fix, and it stays current. If something's missing, you don't have to fork to get it — file a ticket at natebuildsai.com and it'll be weighed for the roadmap (heard and considered, not a guarantee every request ships).
Documentation, schemas, and examples here are licensed under the Apache License, Version 2.0 — see LICENSE and NOTICE. The Apache license does not grant rights to the project name or branding — see TRADEMARKS.md.