Which agents work with it?

Any agent that reads skills from a directory. Claude Code, Codex and Gemini CLI read it natively; Cursor and VS Code connect over MCP.

The voice runs on your own machine, so there is nothing to meter. The first run downloads it, about 340 MB.

Yes. The skill is MIT-licensed and the authoring format is public.

Create

Ollama v0.40.14 runs your models at their own temperature through the OpenAI API

1 hour ago

@ollama Subscribe

Ollama v0.40.14 stops overriding a model's own temperature and top_p on its OpenAI-compatible API, so apps built on an OpenAI SDK get the settings each model ships with. Claude Code launched through Ollama gets each cloud model's output limit, up to 128,000 tokens, and the Responses API keeps reasoning, images and refusals in resent chats. https://github.com/ollama/ollama/releases/tag/v0.40.14 https://github.com/ollama/ollama/compare/v0.40.13...v0.40.14

Ask about this presentation

Chapters

0:00Ollama’s OpenAI API now runs each model at its own temperature
0:18Claude Code on Ollama cloud models can write answers up to 128,000 tokens long
0:33The Responses API keeps reasoning, images and refusals when an app resends a chat
0:48Claude Code web search comes back full, and the ChatGPT app connects cleanly
1:00Update if your apps reach Ollama through an OpenAI SDK

Show transcript

When an app skips temperature and top P, Ollama's OpenAI-compatible endpoints now use the model's own values, matching Ollama's chat API. LFM3 runs at 0.1 instead of 1.0, and anything your app sends still wins.
Slide text

Ollama
NOV 3, 2026

Ollama’s OpenAI API now runs each model at its own temperature

LFM3 before, temperature sent when an app sends none
1.0

LFM3 now, its own packaged temperature
0.1

Values an app sends still win, including 0.

Ollama release notes, Nov 3 · PR #19432
Ollama launch Claude now gives Claude Code each cloud model's own output limit, up to 128,000 tokens, so long-thinking models stop getting cut off at 32,000. A limit you set yourself still wins.
Slide text
Ollama
NOV 3, 2026

$ ollama launch claude

Cloud models only. An output limit you already set still wins.

Ollama release notes, Nov 3 · PR #19410
Apps that resend a conversation through Ollama's Responses API keep its reasoning and images now. An earlier refusal stops breaking the request, and an image Ollama can't load becomes a short note to the model.
Slide text
Ollama
NOV 3, 2026

Reasoning and images stay in earlier turns

A past refusal no longer fails the request

An image Ollama can’t load becomes a note to the model

Ollama release notes, Nov 3 · PR #19192
Claude Code web searches through Ollama's cloud stop coming back empty. Launching the ChatGPT app checks first that your server can route it. And a failed ChatGPT disconnect can now be retried.
Slide text
Ollama
NOV 3, 2026

Web search through Ollama’s cloud stops returning empty replies

Launching the ChatGPT app checks Ollama’s routing first

A failed ChatGPT disconnect can be retried

Ollama release notes, Nov 3 · PRs #19476, #19470, #19336
If your apps talk to Ollama through an OpenAI SDK, I would update today, because your models finally run at the settings their makers chose. Ollama's release notes list every change.
Slide text

Ollama
NOV 3, 2026

Ollama release notesgithub.com/ollama/ollama/releases

github.com/ollama/ollama/releases

Up next