Using the AI Assistant
The assistant lives in a panel docked to the right of the workspace. You open it, type a question, and read the answer next to the thing you were already looking at.
Everything it can see and change comes from your own permissions and project access. See AI Roles and Permissions.
Open the panel
| Way in | What happens |
|---|---|
| The AI button in the workspace header (a sparkles icon) | The panel opens |
| ⌘/Ctrl + I | The panel opens and the cursor lands in the composer, ready to type |
The panel is resizable by dragging its edge, and it remembers its width and whether it was open — per user, so your layout is yours.
Two markers can appear on the AI button:
| Marker | Meaning |
|---|---|
| A small dot | A run finished while the panel was closed. Open it to read the answer |
| A different indicator | The AI Service is not ready. See Troubleshooting |
The button is absent rather than greyed out when your role lacks Use AI, or when AI is not licensed and enabled. There is nothing to click and nothing you have missed — ask an administrator. See AI Roles and Permissions.
What the panel holds
The header has three controls: new chat, conversation history, and close.
The composer is where you type. Its placeholder tells you what it accepts:
Ask me a question… / for skills, @ to reference, Enter to send
The composer footer holds four controls, left to right:
| Control | What it does | Detail |
|---|---|---|
| Attach (paperclip) | Adds a file or an image to the message you are about to send | Attachments and Context |
| Mode dropdown | Decides whether this conversation may change anything, and how the answer is shaped | Modes and Project Scope |
| Project scope selector | Tells the assistant which project "this project" means | Modes and Project Scope |
| Model picker | Chooses which model answers this one message | Choosing a Model |
The header controls map to the rest of this section too:
| Control | What it does | Detail |
|---|---|---|
| New chat | Starts a fresh conversation. The mode resets, and the scope starts at the project you are working in | Conversations |
| Conversation history | Reopens an earlier conversation, with its mode and scope as you left them | Conversations |
Your first question
Worth doing once, because it shows you what the answer looks like.
- Open a project and a workflow you already know.
- Press ⌘/Ctrl + I.
- Leave the mode on Default and the model on Auto. Both defaults are the right choice for a question.
- Check the project scope selector shows the project you are in. On a new chat it will.
- Type
what does this workflow do?and press Enter.
Watch it read the workflow, then explain it. Nothing was changed — Default cannot change anything.
The Prompt Library has prompts for diagnosing a failing node, planning a change, and building an interface, with the mode to use for each.
Reading the answer
An answer has three parts:
| Part | What it is |
|---|---|
| The reply | The explanation, plan, or account of what was changed |
| N steps of evidence | A collapsed list of what the assistant actually read to get there. Expand it to check its working |
| Links | Anything it names or changed is clickable — the workflow, the node, the document. Click through to the real thing |
The end of a run shows the tokens used and the time taken.
Running a turn
While a run is in flight the Send button becomes Stop.
- Stop is honoured at the next step boundary, never in the middle of a tool call. Work already done is kept, and the conversation shows "Stopped."
- Every run is bounded: at most 24 steps, 16 tool calls, and 120 seconds. Reaching a limit stops the run and reports what it accomplished, rather than failing silently.
- When a conversation gets long, earlier turns are condensed into a summary and the chat says so once.
If a run keeps hitting its limits, the request is doing too much in one turn. Ask for the first part, then the next — a conversation keeps its context between turns.
Verify it worked
- The AI button is in the workspace header, and ⌘/Ctrl + I opens the panel with the cursor in the composer.
- The composer footer shows a mode, a project, and a model.
- A question about something you can already see gets an answer, with an N steps of evidence list you can expand.
- Closing and reopening the workspace brings the panel back at the width you left it.
If it didn't work
| Symptom | Cause | Fix |
|---|---|---|
| No AI button in the header | Your role lacks Use AI, or AI is not licensed and enabled | Ask an administrator to tick Use AI on your role — see AI Roles and Permissions |
| "The AI Service is unavailable" | The local AI Service is not ready yet, or has stopped | Wait a moment and send again — Grid restarts it. If it persists, see Troubleshooting |
| "No AI model is available" | No provider is configured, or every model is disabled | An administrator needs to add a provider and enable a model — see Model Providers |
| The assistant says it cannot make changes | The mode was Build but your role holds no editing permission, or the conversation has no project in scope | See Modes and Project Scope |
| The panel opens but the composer is empty of controls | The panel is dragged too narrow | Widen it by dragging its edge |
| A run stops early with a summary of what it did | It reached a step, tool call, or time limit | Ask for the remainder in the next message |
Next
- Modes and Project Scope — what a conversation is allowed to do
- Choosing a Model — Auto, cost bands, and when to override
- Attachments and Context — files, images, and what the assistant already knows
- Approvals — reviewing a change before it happens
- Conversations — history, new chats, and what is remembered
- Prompt Library — prompts that work
- Project documents — answer from your own specs and guides
- AI Assistant — the section overview
- Troubleshooting