Your first session

1. Connect a provider

Open Settings -> Providers. Connect a built-in provider, add a custom OpenAI-compatible or Anthropic-compatible provider, or use a local no-auth provider such as Ollama or LM Studio.

Built-in provider rows include OpenAI, ChatGPT Plus/Pro, OpenRouter, Anthropic, Gemini, GitHub Copilot, Ollama, LM Studio, OpenCode Go, and OpenCode Zen.

2. Pick a model

Use the model selector in the chat input. Built-in providers can fetch remote model lists. Custom providers use the models you add under that provider. Vision-capable models show an eye indicator and enable image attachments.

3. Open a project

In Projects, click Open project and choose a folder. AgentsOS stores metadata about the folder but does not copy or move the project.

4. Start a thread

The project sidebar lets you create New Thread for the native runtime. If external agents are detected, the same menu also includes runtime-specific threads such as Codex, Claude Code, or OpenCode.

5. Send a message

Type in the chat editor and press Enter. Use @ to mention files, drag files from the tree into the editor, or use /plan-mode when you want a plan before implementation.

6. Review tool calls

Safe Mode is the default permission level. Most routine reads, file patches, note/task updates, and non-dangerous commands run without interruption, while risky tools and dangerous shell commands ask through the approval overlay.

Next steps

• Use the Files tab to inspect code and mention exact lines back into chat.
• Toggle the terminal with Ctrl + </code> or <code>Cmd + .
• Connect MCP servers when the agent needs tools beyond the built-ins.
• Create a task project and dispatch a task when you want a dedicated worktree-backed agent session.