Get Started

Chat With Models and MCP Context

Use AI Chat to connect a language model provider, select a model, add files or saved context, and run workflows through prompts, resources, skills, and MCP tools.

Updated: May 13, 2026

What you need

  • Provider: LM Studio or a configured external provider.
  • Model: a loaded or available model selected in the sidebar.
  • Context: optional files, prompts, resources, skills, or MCP integrations.

Quick start in 5 steps

  1. Open AI Chat in the sidebar.
  2. Click Connection Settings, choose a provider, set the server URL and token if needed, then save.
  3. Click Connect LLM, refresh models if needed, and select a model.
  4. Type a message, or insert a skill, prompt, resource, or file attachment from the input buttons.
  5. Send with the arrow button or Command-Enter, then use transcript search or export when you need to preserve the result.

Connect the provider before sending

The sidebar starts with Connections. Use the provider picker, connect button, refresh button, and Connection Settings window to prepare the model list.

  • Provider: choose LM Studio or an external provider such as OpenAI, Anthropic, Gemini, Mistral AI, or Hugging Face when available.
  • Server URL: include the port for local services such as LM Studio.
  • Authentication Token: leave it blank only for endpoints that do not require authentication.
  • System Prompt: optional text sent after each successful connection.
AI Chat starts the MCP server automatically when the chat view appears. Use the play or stop control in the toolbar if you need to restart it while testing integrations.

Add context from the input area

The input area supports direct text, drag and drop, file attachments, prompt templates, resources, and skills.

  • Select Skill: choose a saved skill with Command-Control-S.
  • Add files: attach files with Command-Control-A or drag and drop into the input.
  • Insert Prompt: add a saved prompt with Command-Control-P.
  • Insert Resource: add saved resource content with Command-Control-R.
File attachments and dropped folders may require Security Scopes. If the app cannot read a path, add that file or folder in Settings before retrying.

Common errors and quick fixes

  • Send button is disabled: connect the provider, select a model, and make sure the input has text or a selected skill.
  • No models found: refresh the model list, check the provider URL, or load/download a model in LM Studio.
  • Authentication fails: reopen Connection Settings, update the token, and reconnect.
  • Attachment fails: grant Security Scope access to the file or folder and add it again.

Next steps