Skip to main content
Axosun is currently unavailable, and its return to service is not planned in the near future.
The work of Axosun is organized in sessions and launches. The session associates a series of launches with a single resource (node, Flow, etc.); the launch is a single call to the assistant: from collecting context to the report or the action performed.
Required rights: session of the panel administrator; for confirmation of actions - permissions according to policy Axosun.

Sessions

  • A manual request for a report always opens new Session: An open session on the same resource closes automatically and remains in history.
  • The validation continues the current session, followed by an additional run result.
  • The session closes automatically when the terminal action (restart, node withdrawal from rotation, configuration application, firewall switching) is successful and was the only launch proposal.
  • Closed sessions are available to read in history: panel Axosun tabs Sessions and History..

Launches.

The launch goes through stages: in queue → context collection → analysis → completed / error / canceled. Streaming Progress. In the chat, the launch is displayed in real time: the current phase, the response text as it is generated, the final report and action cards. In agent mode, additional tool calls, their results, permission cards and short assistant notes about the progress of work. The connection is supported by service signals so that proxies do not break it during long checks.
If the browser is disabled in the middle of the launch, the result is still saved – update the page, and the launch will appear in history.

Cancel launch

Click Stop in the chat - the unfinished launch goes to the status “canceled.” If the launch was waiting for a response to the resolution card, the cancellation will be noticed in about 2 seconds.

Confirmation and rejection of actions

Actions requiring confirmation are shown with cards, either in a report (classic mode) or directly in a launch pause chat (agent mode). The confirmation performs an action and triggers an automatic verification of the result; the deviation is returned to the assistant, and he completes the work without action. The model of rights.

Simplified explanation

For a completed run, you can request a simplified retelling - the assistant will re-explain the result in simpler language in a separate view, without changing the original report.

Limits of agent regime

Agent launch is limited in number of steps and time: no more than 8 steps and 6 minutes to launch. When the limit is exhausted, the assistant saves a report with a note about the reached limit - the result is not lost. Waiting for a response to a resolution card is limited separately to 15 minutes, regardless of the launch budget: a pause on the operator’s decision does not “eat up” the analysis time.

Reply to clarification

Axosun can ask a clarifying question with variants of the answer. The answer continues the original query with its context; it’s not a new independent task. As you run, distinguish between the requested action, the command sent and the confirmed tool result. The presence of the answer text does not in itself prove a change in Flow or node.

Verification

1

Run a test run.

Ask the assistant a diagnostic question and wait for the report.
2

Look at the history.

Panel → Axosun → tab History: the launch should be displayed with the status “completed”.
Launch seen in story with final status, report opens.

Rollback

  • Uncompleted start can be replaced by **** button Stop.
  • The proposed actions can be rejected by **** - the rejected actions are not performed.
  • The open session can be closed manually by by; it will remain in history for reading.
The history records themselves are not deleted - this is part of the audit trail, see Operations Journal and Audit.

Limitations and risks

  • Agent mode works in streaming chat; launches without streaming are performed in classic mode.
  • If the connection broke while the launch was waiting for the permission card, the “on-site” resumption is not supported: confirm the action in the usual way from the report – this will start a new start.

Typical errors

Wait for completion or press Stop. If the problem recurs, check the model profile with the button Check the connection - see Connection LLM provider.
An LLM provider error in the middle of the stream captures the launch as a “mistake” to keep the chat from hanging. Check the provider’s key and availability, then repeat the request.
The result is saved regardless of the connection. Update the page and open the tab History.