feat: add LiteLLM as provider preset

[RheagalFire]

Summary

Add LiteLLM as a provider preset, giving users one-click access to 100+ LLM providers (OpenAI, Anthropic, Azure, Bedrock, Vertex, Groq, Ollama, etc.) through the LiteLLM proxy.

Since LiteLLM proxy speaks OpenAI-compatible protocol, it reuses the existing openai provider with zero new provider code - just a preset entry alongside vLLM, LM Studio, Groq, etc.

Changes

• src/api/providers/types.ts - added litellm preset to PROVIDER_PRESETS (baseUrl: http://localhost:4000/v1, providerId: openai, isLocal: true)

Tests

Tests (18/18 pass):

npx vitest run src/api/__tests__/provider-types.test.ts

 Test Files  1 passed (1)
      Tests  18 passed (18)
   Duration  157ms

All preset validation tests pass (unique IDs, valid providerIds, local presets have no placeholder, unique ports).

Live E2E (LiteLLM proxy -> Claude Sonnet via Azure AI Foundry):

Model discovery via /v1/models:

Models: anthropic/claude-sonnet-4-6

Chat completion via /v1/chat/completions:

Response: 4
Model: claude-sonnet-4-6
Tokens: prompt=20, completion=5

Streaming via /v1/chat/completions?stream=true:

Streamed: Hi|!

All three endpoints verified end-to-end through the same OpenAI-compatible path that the existing OpenAIProvider class uses.

Example usage

In the app: Settings > Providers > select "LiteLLM" from the preset dropdown. Models are auto-discovered via /v1/models.

Start the LiteLLM proxy:

pip install litellm
litellm --model anthropic/claude-sonnet-4-6
# Proxy runs on http://localhost:4000

Python usage example (LiteLLM SDK):

import litellm

response = litellm.completion(
    model="anthropic/claude-sonnet-4-6",
    messages=[{"role": "user", "content": "Hello!"}],
    drop_params=True,
)
print(response.choices[0].message.content)