-
Notifications
You must be signed in to change notification settings - Fork 6
Using FutureOS
This is a tour of the app: the three-panel layout, Chat vs. Workspace, talking to the agent, the approval mechanism, and how to check the agent's work.
FutureOS is organized into three columns:
- Left — navigation. From top to bottom you'll find: New Chat, Models (a shortcut into settings), Skills, your Workspaces (each expands to its conversations), your Chats, and Settings at the bottom. You can collapse the left panel to give the conversation more room.
- Center — the conversation. Your messages, the streaming reply, plans, tool activity, command previews, errors, and approval cards. The input box is fixed at the bottom.
- Right — the context panel. See what the agent is doing (Runs / Review / Artifacts). It's collapsible — open it when you want to check the work, hide it when you don't.
Each conversation is its own independent agent session — they don't interfere with each other.
| Chat | Workspace | |
|---|---|---|
| How you create it | New Chat — just start typing | Workspace — open a folder on your computer |
| Best for | Quick questions, one-off tasks | Real projects tied to a folder |
| Right panel shows | Runs and Artifacts | Runs and Review |
- A Chat is the fastest way to ask something. Its work area is temporary.
- A Workspace binds a real folder, so the agent reads and changes files there — and the Review view shows exactly what changed.
You can rename, pin, or delete conversations from the menu next to each one in the left panel.
- Type in the input box and press Enter to send. Use Shift+Enter for a new line.
- The model selector and thinking level control sit inside the input box — you can switch models per conversation.
- Attach up to 4 images per message with the paperclip button, by pasting, or by dragging files onto the box.
- While a reply is streaming, the send button becomes a stop button so you can interrupt.
This is the heart of FutureOS. When the agent wants to do something with real-world consequences — read or write a file, run a shell command, delete files, or write outside the workspace — it stops and shows an approval card just above the input box. It waits for you, with no timeout.
The card tells you exactly what's being requested — the command to run, the files to write (with a preview), or the paths involved. Then you choose:
- Allow once — let this one action through and continue.
- Deny — cancel the action. The agent is told, so it can adjust and try a different approach.
- Allow in this workspace / this chat (when offered) — save a rule so similar actions in this project don't ask again. You can edit the path pattern before saving.
Keyboard: Cmd/Ctrl+Enter approves, Esc denies (or closes the rule editor first).
You can set how much the agent asks in Settings → General, or from the shield control in the input box:
- Manual — prompts before file reads and writes; read-only commands run automatically.
- Sandboxed (macOS only) — commands run inside the macOS sandbox; file operations still prompt.
- Unrestricted — no prompts and no sandbox; everything runs. Use with care.
Open the context panel on the right and pick a view from the dropdown at its top.
Every background program the agent runs shows up as a card with the real command, its status, and a running/finished count. You can:
- Inspect a run to see its details.
- Terminate a running program.
- Clear finished to tidy up.
For a Workspace, the Review view shows the file changes made in the project: the list of changed files, change types (added / modified / deleted / renamed), and per-file diffs. When the folder is under version control, you can also switch to a "Last run changes" view to see just what the most recent run changed.
For a Chat, the Artifacts view collects the outputs the agent produced — reports, summaries, tables, and files. You can preview them, copy their contents, and export or open the originals. You can also upload a file into Artifacts yourself.
- Quick Start — the fast path to your first answer.
- Settings — providers, models, and approval mode.
- Skills — capability packs the agent uses automatically.
FutureOS · runs on macOS and Windows · Download · Report an issue
Getting started
Using the app
Command line (advanced)
Integrations
Help