[RheagalFire]

Issue for this PR

Closes #29935

Type of change

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

What does this PR do?

Adds LiteLLM as a plugin. Sets LITELLM_BASE_URL (and optionally LITELLM_API_KEY) to connect. Models are synced from the proxy's /v1/models endpoint and enriched with per-model metadata from /model/info.

Files changed:

• src/plugin/litellm/litellm.ts: Plugin with provider.models hook (model sync) and auth hook (API key)
• src/plugin/litellm/models.ts: Model fetching from /v1/models + /model/info metadata enrichment
• src/plugin/index.ts: Registered LiteLLM in internal plugins
• src/provider/provider.ts: Removed built-in provider code, added generic plugin provider seeding
• packages/core/src/provider.ts: litellm provider ID
• sprite.svg / types.ts: Provider icon
• en.ts / settings-providers.tsx / dialog-select-provider.tsx: UI strings and notes

Env var	Description
LITELLM_BASE_URL	Base URL of the LiteLLM proxy (e.g. http://localhost:4000)
LITELLM_API_KEY	API key / master key for the proxy (optional if proxy has no auth)

Related: #29308, #6231, #14468

How did you verify your code works?

1. Started LiteLLM proxy with config (claude-sonnet-4-6 + claude-opus-4-6 via Azure AI Foundry)
2. Set LITELLM_BASE_URL=http://localhost:4000 env var
3. Ran TUI (bun dev), LiteLLM models auto-discovered from /v1/models and appear in model picker
4. Verified /model/info metadata enrichment (token limits, costs)
5. Sent messages with streaming, works
6. Switched between models (sonnet/opus), works
7. Full test suite passes (2979 pass, 0 fail)
8. Typecheck passes across all 21 packages

Screenshots / recordings

ceb2-3aa7-47fd-b86f-e27af97bfe24-2.mp4

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:

Potential Duplicate/Related PRs Found:

1. PR feat(opencode): add LiteLLM provider with auto model discovery #14468 - feat(opencode): add LiteLLM provider with auto model discovery

   • feat(opencode): add LiteLLM provider with auto model discovery #14468
   • Why related: This appears to be an earlier or alternative implementation of LiteLLM provider integration with auto model discovery. Current PR feat(opencode): add LiteLLM plugin integration #29937 is likely updating/completing this feature (PR feat(opencode): add LiteLLM provider with auto model discovery #14468 is mentioned in the "Related" section).

2. PR fix: inject cache_control on content blocks for openai-compatible proxies to Anthropic backends (Bifrost, LiteLLM, Databricks) #25985 - fix: inject cache_control on content blocks for openai-compatible proxies to Anthropic backends (Bifrost, LiteLLM, Databricks)

   • fix: inject cache_control on content blocks for openai-compatible proxies to Anthropic backends (Bifrost, LiteLLM, Databricks) #25985
   • Why related: Involves LiteLLM proxy handling and OpenAI-compatible protocol features that may be relevant to the integration.

3. PR feat(opencode): add LLM provider fallback chain #26292 - feat(opencode): add LLM provider fallback chain

   • feat(opencode): add LLM provider fallback chain #26292
   • Why related: Provider integration feature that may have overlapping scope with adding a new gateway provider.

The most significant is PR #14468 - verify if it's the predecessor/duplicate of the current work or if it's a separate issue being addressed by PR #29937.

[RheagalFire]

cc @adamdotdevin would love your review on this one.

[RheagalFire]

Hey folks, any update here
cc @Hona @Brendonovich

[KMontag42]

would be great to have this merged into a release

[el-AI-wei]

This PR got held previously up because it never got reviewed. Now it's blocked because it's stale and conflicts with dev branch and likely needs to be redesigned or ported for the current provider/V2 architecture. So, either this PR needs to do that and somehow get an actual review this time or someone needs to make a new PR.

[thdxr]

This branch has been upgraded to V2. We're no longer taking PRs for V1. If you think this is still relevant, please port it over and open it for V2.

— from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊