Desktop App
Local AI

Local AI in the desktop app

The mdedit.ai desktop app can use local text models through Ollama or OpenAI-compatible endpoints such as LM Studio, llama.cpp server, LocalAI, or vLLM.

Local AI is available only in the Electron desktop app. The web app does not connect to localhost or private network model servers.

Requirements

  • A paid mdedit.ai plan that includes desktop local model access.
  • The mdedit.ai desktop app.
  • A running local model server.
  • At least one installed chat model.

Local model inference does not use mdedit.ai hosted AI credits. Plan access still controls whether local model settings are shown.

Ollama setup

  1. Install Ollama and start it.
  2. Pull a chat model, for example ollama pull llama3.1.
  3. Open mdedit.ai desktop.
  4. Go to AI Settings.
  5. In Local models, choose Ollama.
  6. Use http://127.0.0.1:11434 as the endpoint URL.
  7. Select Save and test.

The test calls Ollama's local model list endpoint and saves the first detected model as the default if no default model is already set.

OpenAI-compatible endpoints

Use this option for local servers that expose OpenAI-style APIs.

  1. Start your local server.
  2. Confirm it exposes a models endpoint, usually /v1/models.
  3. Open mdedit.ai desktop.
  4. Go to AI Settings.
  5. In Local models, choose OpenAI-compatible.
  6. Enter the local base URL, for example http://127.0.0.1:1234/v1.
  7. Enter a default model name if the server does not return one from /models.
  8. Select Save and test.

OpenAI-compatible local endpoints do not require a cloud OpenAI API key. If your local gateway requires its own token, store that in the gateway rather than in mdedit.ai local model settings.

Security boundary

Desktop local endpoints are accepted only when they use:

  • localhost
  • 127.0.0.1 or another private IPv4 address
  • ::1
  • host.docker.internal
  • .local hostnames

Public cloud endpoints such as https://api.openai.com/v1 are rejected in local model settings. Use provider key settings for cloud providers.

The desktop app stores local endpoint configuration on this device. Desktop provider API keys, when used for BYOK desktop routes, are encrypted with Electron secure storage when the operating system provides a secure backend.

Privacy and history

Local model inference runs against the endpoint you configured on this device or private network. Your prompt and document context are sent to that endpoint for the AI turn.

Conversation history can still sync with mdedit.ai cloud history for the document. Do not use local AI for content that must never be stored in cloud-backed history until a local-only history setting is available.

Supported capabilities

Local models currently support text chat, writing, and rewrite-style assistant actions.

Use a cloud model for:

  • image generation
  • animations
  • long-running agent tasks
  • provider-hosted tools

If you select a local model for an unsupported workflow, mdedit.ai should keep the action unavailable or show a clear unsupported-capability message.

Troubleshooting

No models found

Install or pull a chat model in your local runtime, then run Save and test again. For Ollama, ollama list should show the model.

Endpoint unavailable

Make sure the local model server is running and that the endpoint URL includes the right port and path. Ollama usually uses http://127.0.0.1:11434; LM Studio commonly uses http://127.0.0.1:1234/v1.

Missing model

Enter a default model name in AI Settings or run the endpoint test so mdedit.ai can discover available models.

Endpoint rejected

Local model settings accept only localhost, .local, Docker host, loopback, or private network addresses. Public AI APIs belong in provider key settings, not local model settings.

Web app does not show local models

This is expected. Local models are a desktop-only feature because the web app does not open localhost or private network AI connections.

Related

Local Markdown filesOverview