AI Assistant

The Axion AI Assistant is the permanent chat tab. It streams responses token-by-token, supports a message queue, and shows a live plan while autonomous work runs.

Sending messages

Type in the composer and either press the send button or Ctrl+Enter (plain Enter inserts a new line).

Queueing while the agent works

While the agent is streaming, the send button turns amber. Clicking it (or Ctrl+Enter) queues the message instead of dropping it:

Queued messages appear in the queue panel above the composer: N messages queued — sent automatically when the agent finishes.
Each queued message has three actions:
- ➤ Send now — jumps the queue (sent as soon as the current run ends)
- ✎ Modify — returns it to the composer for editing
- ✕ Delete — removes it; it will never be sent
↩ Undo last takes the most recent queued message back into the composer.

Streaming with your provider

Responses stream from your enabled fallback tier — the router sends the request to the tier's model, endpoint, and API key automatically. See Fallback Tiers to configure providers.

The plan while it works

When the agent executes a DAG pipeline, the Plan & Tasks tab (first tab of the diagnostics panel) shows:

The plan goal (short header)
One line per step with an LED: gray pending, orange running, green complete, red failed
A progress line like 3 / 7 steps complete

Asking you questions (question cards)

When the agent needs a decision from you, it can ask right in the chat. Its reply ends with a question card:

Numbered options — each with a short description; the agent marks the one it recommends
Free-text box — type your own answer instead
Skip — dismiss the question

Click an option (or send typed text) and the conversation continues immediately; the card collapses to Answered: … or Question skipped.

The Axion Agent

The default agent driving all of this is Axion Agent — see Axion Agent for its full capability set, default MCP toolkit, and how to uninstall it.

The row of icons under the prompt box is laid out in three groups:

Position	Controls
Left	Attach (files/media/symbols) and Skills & Agents (toggle which skills and agents are injected into the prompt).
Centre	Auto Continue, Auto Router, and Fallback - each opens a flyout of presets plus a link to its full configuration view.
Right	Context / Compact, Optimize prompt, and Send (or the amber Queue button while the agent is busy).

Context / Compact

The Context / Compact icon on the right does two jobs:

Hover it to see context usage at a glance.
Click it to open a popup with:
- a usage bar and the breakdown (system prompt, attached context, conversation history),
- Compact & compress this session now - runs the active compactor over your pinned files and open tabs immediately, and folds older conversation turns into a single summary note. The popup reports exactly how many tokens were saved.
- Open Context & Budget… - jumps to the full Context view.

Compact vs. usage. Context usage just shows you the numbers. Compact actually changes the session right away, making the next prompt smaller.

New Session

The New Session icon sits in the chat header, immediately to the left of the active DAG label. Click it to clear the conversation and start fresh.