Local Models
Local providers keep note content off hosted APIs. The current local path covers Ollama with the native ollama transport and LMStudio through the OpenAI-compatible transport. No API key is required for either path, but the local server and model must already be running.
This is part of the Obsidian AI Knowledge Management Guide.
Setup
For Ollama, install Ollama, pull the model you want, start the server, and add an Ollama provider profile in Notemd. For LMStudio, load a model in LM Studio's local server and add an LMStudio provider profile.
Local models are operational dependencies, not just settings. Notemd can only call what the local process exposes, so verify the local model outside Obsidian before debugging the plugin.
Endpoint And Authentication
| Provider | Base URL | Default Model | API Key | Notes |
|---|---|---|---|---|
| Ollama | localhost:11434/api | llama3 | Not required | NDJSON streaming; model discovery via /api/tags |
| LMStudio | localhost:1234/v1 | local-model | Not required | Sends "EMPTY" as bearer token; OpenAI-compatible |
Ollama uses the native /chat path below its Base URL. LMStudio follows the OpenAI-compatible chat/completions route and expects a model that is loaded in the local server.
Model Discovery
Ollama uses the ollama-tags discovery mode and lists locally pulled tags. LMStudio uses OpenAI-compatible model discovery when its server exposes /v1/models.
If discovery returns nothing, that is usually a local-server state issue: the model is not pulled, not loaded, or the server is bound to a different host/port than the profile.
Troubleshooting
- Connection refused: start Ollama or LM Studio and confirm the configured port.
- Model not found: pull the Ollama tag or load the LM Studio model before testing.
- Slow output: lower context size, use a smaller local model, or route heavy tasks to a hosted provider.
- Empty or malformed responses: confirm the selected provider profile matches the local server protocol; Ollama and LMStudio are not interchangeable.
When To Use
Use local providers for private vaults, offline work, or cheap repetitive tasks. Do not assume local is automatically better for research synthesis or long reasoning; model quality and hardware limits are the tradeoff.
Next Steps
- LLM Providers -- Full configuration and per-task model setup
- Troubleshooting -- Fix connection-refused errors when the local server is not running