Skip to main content
When Ankra AI does not do what you asked, the portal almost always says why, in one sentence, where you asked. This page lists those sentences by symptom, with what causes each and what to do. Start with the symptom you see.

Before you start

  • Note where you asked: Ask Ankra in the portal, the CLI, Slack or Teams, or a pull request. The checks differ.
  • Some fixes need a role that manages Ankra AI, such as owner or admin: resuming a stopped Ankra AI, adding a model key, changing a connection. Where a fix needs one, the portal says who can make it.
  • The CLI checks below need the Ankra CLI, logged in to your organisation:
ankra ai status shows the active AI provider and the configured credentials; ankra ai autonomy status shows whether autonomous actions are paused or all AI is stopped.

Ask Ankra is missing, or there is no Mode switch

See AI Assistant for who sees what.

The question was refused

A question that does not reach Ankra AI is answered in the panel with one of these sentences. A person who can fix it also gets a button to the page that does. When the allowance is used up and Ankra has turned on its DevOps AI model for your organisation, an answer can carry the notice Your organisation’s monthly AI allowance is used up, so this answer comes from the DevOps AI model by Ankra, which has its own monthly allowance. The answer is real; it comes from a different model.

The answer stopped before it finished

In the CLI, a one-shot ankra chat that ends on a pause exits with code 9; add --wait to follow the conversation through its pauses. See ankra chat.

A change was refused or never ran

What Ankra AI may change without asking is set in five places; What it can change without asking lists them.

The model is not configured

Slack, Teams or a pull request gets no answer

  1. Open Ankra AI → Gateway. Each surface shows Connected or Partly connected; one without a badge is not set up.
  2. Press Set up the gateway (or Review the setup) and open the step for that surface. What Ankra found lists what is connected, and the step’s state says what is missing: Not set up, Partly set up, Could not check or Needs an administrator. See Set up the AI Gateway.
  3. For Slack and Teams, check that you linked your own account under Your account → Connected accounts: Ankra AI answers with the access of the person asking.
  4. For a pull request, check that AI review (for automatic reviews) or Replies to mentions (for @ankraai) is on for that repository under Ankra AI → Gateway connections, and for an application under Applications → <your application> → Settings → Who runs it. With Replies to mentions on, a mention of @ankraai is never filtered by branch scope, draft state or the review budgets. See AI code review and Application AI Settings.
AI Connections has the steps for a Slack or Teams app that does not answer.

A run stopped, or a ticket is not worked

Ankra AI → Activity (/next/ai/agent-runs) lists every agent run. Filter it to Failed or Waiting for a person, and open a run: its header says The run failed, Waiting for a person or A plan is ready for review, and Run activity shows each step. From the CLI, read a run in full or its transcript:

Where to look next

  • Needs you (/next/needs-you) - every approval, fix and alert that waits on a person.
  • Ankra AI → Activity - every agent run and its steps.
  • Audit log - what changed in your organisation, and who changed it.
  • Contact support - with the time, the place you asked and the message you saw.