Ollama Generator

Use your Ollama models in LM Studio, with optional API key authentication.

Click Run in LM Studio on the page above to install the adaptor.

For local development, run npm ci && npm test, then npm run dev.

Leave generation options unset to use server defaults. Thinking and tool support depend on the model.

Option	Description
Thinking	Server default, Off, On, Low, Medium, or High.
Context Length	Maximum context tokens per request.
Response Length	Maximum output tokens. Use `-1` for unlimited.
Temperature	Sampling randomness.
Top K Sampling	Candidate token limit. Use `0` to disable.
Repeat Penalty	Repetition control. Use `1` for no penalty.
Top P Sampling	Cumulative probability threshold.
Min P Sampling	Minimum probability relative to the most likely token.
Stop Strings	JSON array of stop sequences, e.g. `["<END>"]`.
Keep Alive	Seconds to keep the model loaded. `0` unloads after a response, `-1` keeps it loaded.
Seed	Sampling seed. Use `-1` for a random seed.
Request Timeout	Maximum seconds per model request.
Maximum Tool Rounds	Maximum tool-call rounds per response.
Context Usage (Beta)	Prompt + output tokens from the last request, with the loaded context limit.

Image attachments are currently unavailable with the adaptor alone.
The SDK provides no API to synchronize native context counters with Ollama.
Ollama's generic stop is shown as EOS; it does not distinguish stop strings.
The SDK cannot hide Prediction Loop warnings (loader.info.activePredictionLoopHandlerWarning).

Use your Ollama models in LM Studio, with optional API key authentication.

Click Run in LM Studio on the page above to install the adaptor.

For local development, run npm ci && npm test, then npm run dev.

Leave generation options unset to use server defaults. Thinking and tool support depend on the model.

Option	Description
Thinking	Server default, Off, On, Low, Medium, or High.
Context Length	Maximum context tokens per request.
Response Length	Maximum output tokens. Use `-1` for unlimited.
Temperature	Sampling randomness.
Top K Sampling	Candidate token limit. Use `0` to disable.
Repeat Penalty	Repetition control. Use `1` for no penalty.
Top P Sampling	Cumulative probability threshold.
Min P Sampling	Minimum probability relative to the most likely token.
Stop Strings	JSON array of stop sequences, e.g. `["<END>"]`.
Keep Alive	Seconds to keep the model loaded. `0` unloads after a response, `-1` keeps it loaded.
Seed	Sampling seed. Use `-1` for a random seed.
Request Timeout	Maximum seconds per model request.
Maximum Tool Rounds	Maximum tool-call rounds per response.
Context Usage (Beta)	Prompt + output tokens from the last request, with the loaded context limit.

Image attachments are currently unavailable with the adaptor alone.
The SDK provides no API to synchronize native context counters with Ollama.
Ollama's generic stop is shown as EOS; it does not distinguish stop strings.
The SDK cannot hide Prediction Loop warnings (loader.info.activePredictionLoopHandlerWarning).

ollama-generator