feat: add LiteLLM as AI gateway provider

[RheagalFire]

Description

Adds LiteLLM as an AI gateway provider, enabling self-hosters to route queries through 100+ LLM providers (OpenAI, Anthropic, Azure, Bedrock, Vertex AI, Groq, Ollama, etc.) via a single LiteLLM proxy. Follows the existing createOpenAICompatible pattern used by ark, sarvam, zai, huggingface, novita, and minimax.

Type of Change

• New feature (non-breaking change which adds functionality)

Related Issues

No existing issues, but this addresses the self-hosting use case where users want to bring their own LLM provider keys through a unified gateway rather than configuring individual provider API keys.

Changes Made

• ai/providers.ts - added litellmProvider via createOpenAICompatible() with configurable LITELLM_BASE_URL and LITELLM_API_KEY env vars, plus one scira-litellm model entry using LITELLM_MODEL env var (defaults to gpt-4o-mini)
• ai/models.ts - added 'litellm' to ModelProvider type union, added litellm entry to PROVIDERS record, added scira-litellm model metadata entry, added 'litellm' detection in getModelProvider() function
• components/ui/form-component.tsx - added LiteLLM bullet train icon to ProviderIcon switch
• components/ui/model-selector.tsx - added LiteLLM bullet train icon to ProviderIcon switch
• ai/__tests__/litellm.test.ts - 11 unit tests for provider registration and model config

Testing

• Tested locally
• Added/updated unit tests
• Added/updated integration tests
• All tests pass

Unit tests (11/11 pass):

$ bunx vitest run ai/__tests__/litellm.test.ts

Test Files  1 passed (1)
      Tests  11 passed (11)
   Duration  483ms

Tests cover:

• LiteLLM provider exists in PROVIDERS record with correct metadata (id, name, icon, hasNew)
• scira-litellm model entry exists with correct metadata (label, provider, experimental, requiresAuth, maxOutputTokens)
• Model description mentions "gateway"
• getModelProvider() returns 'litellm' for scira-litellm and any value containing "litellm"
• getModelProvider() does NOT return 'litellm' for unrelated models (gpt, anthropic)
• getModelProviderInfo() returns correct LiteLLM provider info
• getActiveProviders() includes LiteLLM
• getModelsByProvider('litellm') returns at least one model

Lint clean:

$ bun run lint
(no errors)

Live E2E - local dev with LiteLLM proxy routing to Claude Sonnet 4.6 via Azure Foundry:

Ran full local dev stack (Postgres, Redis, LiteLLM proxy, Next.js dev server). Selected LiteLLM from model selector, sent "hi there", received a response from Claude Sonnet 4.6 routed through the LiteLLM proxy. Screenshot attached below.

Server logs confirmed correct model routing:

Search API: {
  model: 'scira-litellm',
  requestedModel: 'scira-litellm',
  group: 'web',
  isAutoRouted: false
}

Test Environment

• OS: macOS Darwin 25.4.0
• Browser: Chrome
• Node Version: v22.x (via Bun)

Screenshots

Checklist

• My code follows the project's style guidelines
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• My changes generate no new warnings or errors
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published

Additional Notes

Configuration for self-hosters:

Set these environment variables to connect to your LiteLLM proxy:

LITELLM_BASE_URL=http://localhost:4000/v1   # Your LiteLLM proxy URL
LITELLM_API_KEY=sk-your-litellm-key         # Your LiteLLM proxy API key
LITELLM_MODEL=gpt-4o-mini                   # Default model on your proxy

The LiteLLM proxy handles routing to any backend (Azure OpenAI, AWS Bedrock, Vertex AI, Anthropic, Ollama, etc.) so self-hosters get access to all providers without needing individual API keys configured in scira.

Why LiteLLM over adding more individual providers:
LiteLLM acts as a unified gateway. Instead of adding SDK integrations for Bedrock, Vertex AI, Azure, etc. one by one, self-hosters can point scira at a single LiteLLM proxy that handles provider routing, load balancing, cost tracking, and fallbacks across 100+ providers.

[RheagalFire]

cc @zaidmukaddam

[RheagalFire]

@zaidmukaddam do you have any update on this PR?