Skip to content

Every provider starts on its own default model, and you can change it - #155

Merged
antonarnaudov merged 1 commit into
mainfrom
feat/provider-default-models
Nov 7, 2026
Merged

antonarnaudov merged 1 commit into
mainfrom
feat/provider-default-models

Conversation

@antonarnaudov

Copy link
Copy Markdown
Contributor

Each provider now starts on its own default model, and you can change it, in both the extension and the desktop app. This follows up #154 (unreleased) and replaces its "you choose the OpenAI model" behaviour.

  • One table for both products: packages/engine/src/ai/providerDefaults.ts holds each provider's default models. The desktop catalog reads its presets from it, and the extension reads it too.
  • Desktop:
    • OpenAI connections are back on OpenAI's current defaults.
    • A connection still on gpt-4o moves to them when OpenAI retires gpt-4o, as in 2.9.0.
    • Azure starts on the same models instead of gpt-4o.
    • Each model field still lists the provider's models, so you can switch.
  • Extension:
    • An empty model setting now means the defaults of the provider your base URL points at (OpenAI, Groq, OpenRouter, Gemini, Mistral, xAI, DeepSeek, Together, Ollama, LM Studio, Azure). Before, it was gpt-4o for every endpoint.
    • A model of your own in any tier is used as it is, and no default is mixed in.
    • Each card in the panel starts on its provider's default.
    • A detected list never replaces a cloud default. For Ollama and LM Studio, the default becomes a model the server actually has.
    • You choose the OpenAI model; desktop cherry-picks and reverts merge commits #154's one-time migration, which pinned 1.24 users to gpt-4o, is removed.
  • Settings description: the description now says that leaving the fields empty uses the provider's defaults, in all 14 languages.

Tests:

  • apps/extension/test/aiProviderDefaults.test.ts: endpoint detection, default resolution, the request sent, and the panel in headless Chrome.
  • packages/ai/test/retiredDefaults.test.ts: the presets match the table, and gpt-4o connections migrate.
  • npm run verify passes.

🤖 Generated with Claude Code

… change it

One table of each provider's default models (engine/ai/providerDefaults)
now seeds both products. The desktop's catalog reads its presets from it,
and OpenAI connections are back on OpenAI's current defaults (gpt-6-luna,
gpt-6.1-terra, gpt-6.1-sol). A connection still on gpt-4o moves to them
when OpenAI retires gpt-4o, as it did in 2.9.0. Azure starts on the same
models instead of gpt-4o.

In the extension, an empty model setting now means the defaults of the
provider the base URL points at. Until now the settings defaulted to gpt-4o
for every endpoint, so Groq, OpenRouter and local servers were all sent
OpenAI's model. If any tier holds a model of your own, it is used and no
default is mixed in. Each card in the AI panel starts on its provider's
default and lists the models you can switch to. For Ollama and LM Studio,
the default becomes a model the server actually has.

This replaces the unreleased "you choose the OpenAI model" change, including
its one-time migration that pinned 1.24 users to gpt-4o.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@antonarnaudov
antonarnaudov merged commit fc5186b into main Nov 7, 2026
13 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant