feat(opencode): add dynamic model fetching for OpenAI-compatible provider

[shivamashtikar]

Issue for this PR

Closes#12814
Closes#10633

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Adds a shouldFetchModels config option that dynamically fetches available models from the provider's OpenAI-compatible API at startup. This eliminates the need to manually list every model in opencode.json for custom providers.

Comparison with PR #13896:
PR #13896 implements LiteLLM-specific model loading. This PR provides a generic implementation supporting any OpenAI-compatible provider (LiteLLM, Ollama, vLLM, custom proxies, etc.). Both try /model/info first, but this PR adds proper fallback to the standard /models endpoint.

How it works:

• When a custom provider has baseURL configured, models are fetched in the background (unless shouldFetchModels: false)
• First tries /model/info for rich metadata (limits, costs, capabilities)
• Falls back to /models endpoint if /model/info unavailable
• Manual config models always override fetched ones
• Non-blocking - startup proceeds immediately, models merged when response arrives

Example:

{
"provider": {
"my-proxy": {
"npm": "@ai-sdk/openai-compatible",
"options": {
"baseURL": "https://my-proxy.com/v1",
"apiKey": "{env:API_KEY}"
}
}
}
}

To disable:

{
"provider": {
"my-proxy": {
"shouldFetchModels": false,
"options": { "baseURL": "https://my-proxy.com/v1" }
}
}
}

How did you verify your code works?

• bun test test/provider/fetch-models.test.ts - 4 tests pass
• bun test test/session/llm.test.ts - 10 tests pass (no regressions)
• Built standalone executable with --single flag and verified it works

Screenshots / recordings

N/A - no UI changes

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Found a potential duplicate:

PR #13896 - feat(opencode): add auto loading models for litellm providers

• feat(opencode): add auto loading models for litellm providers #13896

Why it's related: This PR appears to implement very similar functionality - automatic model loading for LiteLLM providers. Given that PR #14277 also adds dynamic model fetching for OpenAI-compatible providers (including LiteLLM), these PRs may be addressing the same or overlapping feature requests. You should verify whether PR #13896 was already merged or closed and check if #14277 extends/improves upon it.

[shivamashtikar]

this PR provides compatibility with any open-ai compatible endpoints along with litellm as opposed to #13896 which only handles for litellm

[PratikNarola1]

LGTM

[8x22b]

PLEASE MERGE IT

[espetro]

up! we're looking forward to it!

[shivamashtikar]

Hi @alexyaroshuk@adamdotdevin could you please check this PR and let me know if it can be merged or if you need any more changes in it

[alexyaroshuk]

Hi @alexyaroshuk@adamdotdevin could you please check this PR and let me know if it can be merged or if you need any more changes in it

unit tests failing:

4 tests failed:
(fail) session.llm.stream > sends temperature, tokens, and reasoning options for openai-compatible models [31.00ms]
(fail) session.llm.stream > sends responses API payload for OpenAI models [63.00ms]
(fail) session.llm.stream > sends messages API payload for Anthropic models [94.00ms]
(fail) session.llm.stream > sends Google API payload for Gemini models [109.00ms]

I checked with upstream and confirmed that upstream does not have these failing. Your PR has these failing. You can verify by running bun test from packages/opencode

[shivamashtikar]

Hi @alexyaroshuk , apologies i missed those. I've fixed the test cases and and also added new test cases related to my change, can you please review and merge if everything seems fine

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍

[alexyaroshuk]

Hi @alexyaroshuk , apologies i missed those. I've fixed the test cases and and also added new test cases related to my change, can you please review and merge if everything seems fine

i cannot merge, i simply try to help with review. Looks like the test issue is fixed, but merging is up to Adam

[shivamashtikar]

Sure @alexyaroshuk please help with the review

[github-actions]

Closing this pull request because it has had no updates for more than 60 days. If you plan to continue working on it, feel free to reopen or open a new PR.