Chat, sessions, models, and file controls
Choose a conversation and model, give the assistant context, queue instructions, and navigate its replies.
Step-by-step guidance.
Before you start
- Open the named screen in Rumata. Available controls may depend on your account, permissions, and resource state.
Sessions and conversation navigation
A session is a conversation with its own history. The assistant still works in the same factory.
| Control | What it does |
|---|---|
| Main agent / session row | Open that conversation. The selected row is highlighted. |
| Sub-agent row | Inspect a child conversation created by the assistant. Its indentation identifies its parent. |
| New Session / Create session / plus | Open a dialog for a new conversation. |
| Session name / Create Session | Enter a name of up to 120 characters, then create the session. The name cannot be empty. |
| Rename session / Save Name | Rename a main session. Child sessions do not show the same rename action. |
| Cancel / dialog X | Close the create or rename dialog without applying the pending name. |
| Latest | Jump to the newest messages. |
| Previous user messages markers | Jump back to one of your earlier prompts. |
| Fork | Create a new session branching from that point in the conversation. |
Prompt, mode, model, and queue
Describe the outcome you want in everyday language. Technical users can add exact files, interfaces, and acceptance criteria.
| Control | What it does |
|---|---|
| Take screenshot | When supported by the browser, choose a display source and capture a PNG as a pending attachment. The capture stream stops afterward. Review the image for private data and send the prompt separately; capture alone does not submit it. |
| Message field / Enter / Shift+Enter | Write your request. Enter sends it; Shift+Enter inserts a new line. |
| Plan / Build / agent selector | Choose an available assistant mode. Plan develops an approach for review; Build carries out changes. Available choices depend on the factory’s agent runtime. |
| Model selector | Choose a model from the available provider groups. Connected providers appear first; the image icon identifies models with image input support. |
| Default / Current / Use runtime default label | Check which model is in use when you haven’t selected one explicitly. The label can reflect factory settings or the session’s current model. |
| Send | Submit the prompt, selected references, attachments, and command. While a response is running, a new message can enter the Up next queue. |
| Stop | Ask the assistant to stop its current response. It appears while work is running and the composer is empty. Changes already made are not rolled back. |
| Up next → Move back to composer | Remove that instruction from the pending queue and return it to the input for editing. |
| Up next → Steer | Send a queued instruction into the current response when the runtime supports steering. |
Files, skills, and commands
A reference points the assistant to existing project context. A local attachment supplies a file from your computer.
| Control | What it does |
|---|---|
| @ / Search skills or files… | Open and search the reference picker from the message field. |
| File or skill result | Add the selected reference to the pending prompt. Choosing a reference alone does not send the prompt. |
| Reference chip → X | Remove that reference from the pending prompt. |
| / / Search commands… | Find commands available in the current agent runtime. Read the description before selecting one. |
| Command result / Arguments field | Select a command, then provide the arguments it expects. Send executes the selected command. |
| Command chip → X | Remove the selected command. If the command was typed directly in the message rather than picked, this action clears that typed prompt. |
| Add to chat → Upload from computer | Choose up to six local files for the prompt. Each prepared attachment must fit within 1 MiB; non-image files above that limit are rejected. |
| Attachment chip → X | Remove a file from the unsent prompt. Selecting files does not upload them into the factory’s storage bucket. |
Read replies and tool activity
Expandable activity cards explain what the assistant attempted and what the tool returned.
| Control | What it does |
|---|---|
| Copy below a message | Copy the message text. |
| Tool activity / reasoning / task-list disclosure | Expand or collapse details such as input, output, errors, reasoning text, or the assistant’s task list. Expanding does not rerun the tool. |
| Preview telemetry disclosure | Inspect preview network and console context attached to the conversation. |
| Image attachment | Open a larger image view. Close the dialog to return to the chat. |
| Other file attachment | Open its file link in a new tab. |
| Recognized workspace file link | Open that file in Code, at the supplied position when available. Code access depends on the plan and project setup state. |
| MCPs Connected | Inspect the services available to the assistant and whether each is connected, disabled, failed, or needs authentication or registration. |