feat: enable thinking/reasoning toggle for ollama models

[fbricon]

This PR enables users to toggle thinking for Ollama models that support it. In this recording, you can see qwen3 reasoning is not enabled by default, but setting reasoning : true in the defaultCompletionOptions allows it:

  - name: qwen3:8b
    provider: ollama
    model: qwen3:8b
    defaultCompletionOptions:
      reasoning: true

ollama-models-toggle-thinking.mp4

• shows reasoning toggle if model supports reasoning or config explicitly forces it
• reasoning session state properly resets when changing models
• reasoning status passed as think:true to ollama
• updated docs to mention support of reasoning configuration for ollama models

Hardcoding reasoning capabilities on ollama is tricky: e.g. qwen3:8b has reasoning capabilities but think=true causes an Ollama error if the version pulled into ollama is too old. Pulling a newer version will fix it.

Also worth noting some models, like gpt-oss:20b will just ignore the think:false flag passed to ollama. It supports different levels of thinking, but needs to be set in the system prompt. See https://huggingface.co/openai/gpt-oss-120b#reasoning-levels

---

Summary by cubic

Adds a reasoning/thinking toggle for supported Ollama models. Auto-detects support, shows the toggle in the UI, and sends think to Ollama; docs and schema updated.

• New Features
  • Detects reasoning support (Anthropic, deepseek‑r, or when completionOptions.reasoning is true).
  • Shows the Reasoning toggle only when supported; resets state on model change.
  • Streams reasoning options: sets reasoning and budget for non‑Ollama; sends think to Ollama and skips budget.
  • Updates docs and VS Code schema to note reasoning support for some Ollama models.

[fbricon]

I need to fix those gui tests

[Patrick-Erichsen]

Just a few questions/nitpicks, thanks for the contribution @fbricon ! Appreciate the detailed description and screen recording.

[Patrick-Erichsen]

I believe older models such as 3.5 Sonnet don't support reasoning?

Although I'm a little confused, this file is just for Ollama autodetect, no?

[fbricon]

I just moved that existing check from gui/src/components/mainInput/InputToolbar.tsx

And no, this file seems to handle different models from different providers

[Patrick-Erichsen]

Thanks for making those updates! Lmk if this is good to merge and I'll do so.

[fbricon]

I can't test the Anthropic models so if you can do that and I didn't break anything, then go ahead and merge it

[RomneyDa]

@fbricon have verified this works with e.g. claude 4.1 opus local. Appreciate the contribution!

[sestinj]

🎉 This PR is included in version 1.13.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀