Skip to content

Features and data

Check all of the following:

  • The MCP server is configured and working in the host’s mcp.json.
  • Allow calling servers from mcp.json and authentication are enabled in LM Studio, with suitable token permissions.
  • The ID in Settings → Integrations matches the host’s ID exactly.
  • The integration is enabled in this conversation’s Chat Settings → Integrations.
  • The selected model supports tools, and your request actually calls for them.

Enable Render Tool Calls to inspect activity. If a tool fails, check its own host logs and service authentication. See Integrations.

Check the model’s Vision capability. Use a readable image, stay within four attachments per message, and ask a specific question. If image input fails in LM Studio itself, address the host/model problem first.

Check microphone and speech-recognition permissions, and whether on-device recognition is available. A device-side recognition failure is separate from server connectivity.

For silent replies, check Dialog Mode TTS, the selected Assistant Voice, device volume, and audio output. Try a voice preview in Settings. See Dialog mode.

Check Local Notifications, system permission for LLMster, and Focus settings. Keep the host awake and the network available. Background limits may prevent the app from receiving a completion; an alert cannot arrive for a completion it never received.

Make sure you selected a supported LLMster export rather than a text transcript. For a protected library backup, enter the original password exactly. Wait for validation before proceeding.

If a file is incomplete or damaged, obtain another copy or export it again from the source device. Do not rename or edit the archive’s internal files to bypass validation.

Export All Chats requires a non-empty library and no active responses or tool calls. Model unloading also depends on a reachable server, loaded models, and no conflicting model or title-generation activity. Reset waits for responses, tools, model loads, title generation, and transfers.

Read the status below the action and let work finish. If activity appears stale after an interruption, reopen LLMster and review the conversation state before trying again. See Utilities.

This is expected: app preferences, configured server addresses, authentication, and Keychain data are excluded. Configure those separately. Review Sharing and backups for the transfer boundaries.

For further help, visit Support. Include versions and the error message, but not passwords, API tokens, or private conversation contents.