Introduction

What it is

AgentsOS runs locally and keeps work organized around three top-level sections: Projects, Notes, and Tasks. Projects are where coding sessions happen. Notes are a local Markdown knowledge base. Tasks are a lightweight planning and dispatch surface for turning work items into agent sessions.

The native agent can read files, patch code, run commands, use configured MCP tools, create plan artifacts, maintain live todos, query or update notes and tasks when asked, and call subagents for bounded work. You decide which provider and model to use, and tool calls pass through the active permission mode.

Main screens

• Projects: project list, session threads, chat, file tabs, file tree, file-change history, git toolbar, terminal, and worktree merge controls.
• Notes: folders, Markdown editing, wiki links, tags, backlinks, and full-text search.
• Tasks: task projects, table and kanban views, labels, priorities, due dates, task-agent chat, and code-agent dispatch.
• Settings: providers, custom models, MCP servers, skills, external agents, diagnostics, usage, about, and reset controls.

What you need

• A desktop build of AgentsOS.
• A project folder on disk.
• At least one configured model provider, local model endpoint, ChatGPT Plus/Pro connection, GitHub Copilot connection, or external ACP agent.

How to read these docs

Start with Install, then Your first session. After that, use the navigation by screen. The docs match the native app's current surfaces rather than describing future features.