Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.novacula.io/llms.txt

Use this file to discover all available pages before exploring further.

The AI Assistant is an in-product chat panel that lives in the bottom bar of the Dashboard UI. Use it to ask questions about your fleet, draft config changes, walk through a flow you don’t remember, or get pointed at the right docs page. The assistant is available to every signed-in org role — Org Owner, Org Admin, Org Member. It is not rendered in the Admin Panel.

Open the chat panel

A sparkles-icon launcher sits in the bottom bar on every Dashboard route. Clicking it docks a chat panel above the launcher (~360 × 620 px). A maximise toggle in the panel header expands it to a full-height side panel; the X closes the panel back to the launcher without ending the active thread. The panel is lazy-loaded — the bundle ships only after the first open, so the launcher itself has near-zero cost on initial page load.

Send a message

Type into the prompt input at the bottom of the panel and press Enter or the send button. Responses stream in token-by-token as the assistant composes them. A blank thread shows quick prompts as chips — single-click starters tuned to the part of the product you’re on. They either seed the prompt input (so you can tweak before sending) or kick off an action directly.

Tool calls and confirmations

The assistant can propose state-changing actions on your behalf — apply a draft config patch, run a lifecycle action, mutate a setting. Every such action is gated by an explicit confirmation step in the panel before it runs. You see exactly what’s about to change, with a clear approve / decline choice. The assistant cannot perform writes silently. If the panel doesn’t ask you to confirm, the action is read-only.

Switch and delete threads

A history popover (clock icon) lists every thread you’ve started, newest first. From there you can:
  • Start a fresh thread with the plus button — the previous thread is preserved, not deleted.
  • Switch to an older thread by clicking it. Messages load back into the panel and the input continues the conversation in that context.
  • Delete a thread with the per-row trash button. Confirmation step; if you delete the currently-active thread, the panel resets to an empty conversation.
The active thread id is remembered across navigation and across page reloads, so re-opening the panel returns you to the last conversation you had.

What the assistant can do

  • Answer questions about your fleet using read-only access to your org’s resources (executors, nodes, events, audit log).
  • Walk you through a flow you don’t remember — “how do I rotate an executor API key?” — and link out to the corresponding docs page.
  • Propose configuration changes for review (the confirmation gate above always applies).
  • Reference sources inline when it draws from a specific event, audit row, or doc.

What the assistant won’t do

  • Take state-changing actions without an explicit confirmation in the panel.
  • Operate from the Admin Panel — it is Dashboard-UI only.
  • Persist state outside of threads — each conversation is scoped to its thread.
  • Replace direct UI actions for routine tasks. For one-off mutations the wizard / form is faster.

Privacy and data handling

Threads are stored server-side scoped to your user; threads are private and not visible to other members of your organization. Deleting a thread removes it from the server. The model receives only the information needed to answer — typically the message stream plus tool-resolved context from the active org. For confirmation-gated actions, the precise mutation payload is shown in the panel before you approve.

Permissions

  • Open the panel and chat — Org Owner, Org Admin, Org Member.
  • Approve a state-changing action — your role must already grant the underlying operation. The assistant cannot escalate permissions; an action you couldn’t perform yourself is also one the assistant can’t perform for you.