Skip to main content

Lens Assistant

Use Lens Assistant to ask about clusters and deployments, or request an operation. For example: List failed deployments in my clusters.

Start a conversation​

Lens Assistant currently supports External AI providers only. Internal deployments are not supported yet.

  1. Open Lens Assistant.
  2. Under External AI provider, open Select an AI provider… and choose a configured provider.
  3. Click into the message box when its placeholder shows Message Lens….
  4. Type your question and press Enter.

Choose a model with at least a 32K-token context window; 64K or more is recommended. If the list is empty, ask an administrator to add an OpenAI-compatible provider. Anthropic-type providers are not supported by Lens Assistant.

Start over with a new chat​

  1. Click New chat in the page header.
  2. If needed, reselect the External AI provider you want to use.
  3. Send the first message in the cleared conversation.

Interrupt or interject​

While Lens Assistant is working, the message box stays available.

  • Type a new message and press Enter to stop the current turn and send the new message.
  • Press Esc to stop the current turn without sending anything.

Lens stops scheduling further tool calls and discards the interrupted turn. A tool call that was already running may finish in the background, but its result is not added to the conversation.

Review the assistant's reasoning and tool activity​

  1. After you send a message, expand Thinking to review the assistant's reasoning summary.
  2. Expand Tool call entries to inspect the tool name, arguments, and result.
  3. For Permission required, ask an administrator for the reported permission. Changing pages does not grant access.
  4. For Blocked (unknown tool), use a supported operation. Voice requests cannot make changes.
  5. If an old detail is no longer remembered after a long conversation, restate it or use New chat.

Approve or reject a requested action​

Only destructive, hard-to-undo tool calls (for example deleting a cluster or a deployment) pause for your decision. Read and write actions (listing resources, starting a deployment, and similar) run immediately without a prompt.

  1. When a turn pauses on Needs approval, expand that entry.
  2. Review the request details and arguments.
  3. Select the checkbox I reviewed the arguments above and want to proceed.
  4. Click Approve to continue or Reject to stop that action.
  5. Watch the status change to Approving…, Rejecting…, Approved, or Rejected.

Send a voice message​

A voice-originated turn is restricted to read-only tools; if the assistant tries a write or destructive action in response to a spoken question, Lens blocks it instead of running it.

  1. If your browser supports speech input, choose a language from Voice input language. Language names are displayed in English (for example, Chinese).
  2. Click the microphone button with the tooltip Speak your question (runs locally in your browser).
  3. Speak your request.
  4. Click the microphone again when its tooltip changes to Stop recording.
  5. Wait for Transcribing your voice message…, then let Lens send the transcribed message.

The first transcription downloads a Whisper model through Lens; the composer shows download progress and then the transcription status. Audio processing stays in your browser; model files are proxied because browsers on proxy-only networks may not reach the Hugging Face CDN directly.

Connect an external MCP client​

Lens Assistant uses Lens' built-in MCP tool catalog. To drive the same tools from your own MCP client (MCP Inspector, an IDE agent, …), see MCP integration.