AI chats
An AI chat runs an agent CLI in a conversation Ruimte draws. You answer its questions and approvals in the app or on your phone.
Chat or terminal
Both run the agent CLI on your machine, with its own login.
| Terminal | AI chat | |
|---|---|---|
| Interface | The CLI's own | Messages drawn by Ruimte |
| Questions and approvals | Answered in the terminal | Answered in cards, here or on your phone |
| Plan beside it | No | Yes |
| Tables, maps and pages in replies | No | Yes |
Providers
Chats work with Claude Code, Codex, GitHub Copilot, Grok and Antigravity. Settings, Providers adds more:
- Claude API, with an Anthropic API key
- NanoGPT, with a key for its subscription
- other agents from the public agent registry, under Add provider
Start a chat
- On a canvas, press AltC, or choose Chat under Agents.
- As a view of its own, run New Claude Code chat view (or another CLI) from the command palette.
- Outside a project, press CtrlAltN.
The control beside the composer sets the model, the reasoning effort, the account and the permissions. CtrlShift. raises the effort, CtrlShift, lowers it. A pick becomes the default for new chats of that CLI.
Write a message
These keys open a picker in the composer:
| Key | Picks |
|---|---|
/ | A command of the CLI |
@ | A file of the project |
$ | A skill |
# | A view or node of the project |
A # reference lets the agent read and work in that view or node, as over a line on the canvas. A terminal you reference stays read-only.
- To attach files, click the attach button or drop them on the chat.
- To change which key sends, open Settings, Chats, Send with.
- Alt↑ and Alt↓ bring back messages you sent.
A message you send while Claude Code or Codex works reaches it at its next step. Hold Alt while sending to queue it until the turn ends instead. Turning off Settings, Chats, Send to a working agent right away swaps the two.
Permissions
| Mode | The agent asks before |
|---|---|
| Supervised | Commands and file changes |
| Auto-accept edits | Commands |
| Auto | Actions it does not judge routine |
| Full access | Nothing |
Set the defaults for new agents under Settings, Agents. An agent never gives an agent it opens a wider mode than its own.
Answer a request
A Run command card, or a card for another tool, shows what the agent wants to do. For file changes it shows the diff. Click Allow or Deny, or Always allow when the CLI offers it. To tell the agent why you deny, click Add a reason first.
A question card lists choices. Pick one or more, or write your own under Something else…, then click Answer.
While a request waits, the chat reads Needs you. You can also answer from the Needs you list or your phone; see Needs you.
Control a running chat
| Action | Result |
|---|---|
| Stop | Ends the turn. Shift-click to stop the sub-agents too. |
| Clear | Deletes the conversation, attachments and plans. The chat stays. |
| Restart agent session | Loads new skills, plugins, commands and MCP servers for the next message. The conversation stays. |
| Bookmark message | Marks a message to find it again. |
| CtrlF | Finds text in the chat. |
| Search chats | Searches every chat of the project, from the command palette. |
Settings, Chats, Show what an agent reads and runs hides or shows the file reads and commands in the thread.
Usage limits
When a chat stops on a usage limit, it shows when the limit resets.
- To continue by itself after the reset, turn on Settings, Agents, Continue after a usage limit. A turn that failed on an overloaded model then also tries again later. Resume at reset turns this off for one chat.
- To continue now on another account of the same CLI, click Continue on or pick the account in the account menu.