feat(cli): add OpenRouter support for AI and embedding models

[Neysixx]

Relates to

Improves OpenRouter integration in ElizaOS CLI by adding native embedding support, eliminating the need for users to configure a separate embedding provider when using OpenRouter.

Risks

Low Risk - Well-contained changes following existing patterns

• User Impact: No breaking changes. Existing OpenRouter configurations continue to work unchanged.
• Test Coverage: Unit tests updated to reflect new behavior (embedding support enabled for OpenRouter)
• Configuration: Follows the same pattern as existing providers (OpenAI, Google, Ollama) for both interactive and non-interactive modes
• Validation: Uses existing isValidOpenRouterKey function to validate API key format (sk-or- prefix)

Background

What does this PR do?

This PR adds comprehensive OpenRouter support for embeddings in the ElizaOS CLI:

1. Updates hasEmbeddingSupport to include 'openrouter' as a model with built-in embedding support
2. Adds OpenRouter to getAvailableEmbeddingModels as a fallback option when the primary AI model doesn't support embeddings (e.g., Claude)
3. Implements setup configuration in both setupAIModelConfig and setupEmbeddingModelConfig to handle:
   • Interactive mode: Prompts users for OpenRouter API key
   • Non-interactive mode: Writes placeholder configuration with helpful comments
4. Updates unit tests to reflect that OpenRouter now has embedding support

Technical Details:

• Adds OpenRouter configuration with OPENROUTER_API_KEY, OPENROUTER_EMBEDDING_MODEL, and OPENROUTER_EMBEDDING_DIMENSIONS
• Leverages OpenRouter's unified API access to multiple embedding models (including OpenAI's text-embedding-3-small)
• Reuses existing promptAndStoreOpenRouterKey helper function from get-config.ts

What kind of change is this?

• Features (non-breaking change which adds functionality)
• Improvements (misc. changes to existing features)
• Updates (test updates to reflect new behavior)

Why are we doing this? Any context or related work?

Problem:

• ElizaOS requires embeddings for agent memory, semantic search, and knowledge base functionality
• OpenRouter was only available as an AI text generation provider, not for embeddings
• Users selecting OpenRouter for AI had to configure a second provider (OpenAI, Google, or Ollama) solely for embeddings

Solution:

• OpenRouter provides access to multiple embedding models through its unified API
• This PR enables users to use a single OpenRouter API key for both text generation and embeddings
• Reduces configuration complexity and API key management overhead

Documentation changes needed?

• My changes require a change to the project documentation.
• Documentation partially exists: The .env template in packages/cli/src/utils/get-config.ts already includes OpenRouter environment variables (lines 79-82)

Recommended documentation updates:

1. Add OpenRouter to the list of recommended providers supporting both AI + embeddings in README
2. Document the three OpenRouter environment variables with example values:
   • OPENROUTER_API_KEY=sk-or-v1-...
   • OPENROUTER_EMBEDDING_MODEL=openai/text-embedding-3-small
   • OPENROUTER_EMBEDDING_DIMENSIONS=1536
3. Optional: Add migration note for existing users who can now simplify their configuration

Testing

Where should a reviewer start?

Review Priority:

1. Core logic changes - packages/cli/src/commands/create/utils/selection.ts

   • Line 50: hasEmbeddingSupport now includes 'openrouter'
   • Lines 140-144: OpenRouter added to getAvailableEmbeddingModels

2. Configuration setup - packages/cli/src/commands/create/actions/setup.ts

   • Lines 113-136: setupAIModelConfig OpenRouter case
   • Lines 288-362: setupEmbeddingModelConfig OpenRouter case

3. Test updates - packages/cli/tests/unit/utils/selection.test.ts

   • Line 20: OpenRouter added to models with embedding support
   • Lines 71-72: Correctly categorizes OpenRouter as having embeddings

Detailed testing steps

1. Unit Tests ✅

cd packages/cli
bun run test tests/unit/utils/selection.test.ts

Expected: All tests pass

• ✅ hasEmbeddingSupport('openrouter') returns true
• ✅ OpenRouter correctly identified as model with embedding support

2. Interactive Mode - OpenRouter as Primary AI

elizaos create test-openrouter-primary
# Select: OpenRouter as AI model
# Expected: Should NOT be prompted for embedding model

Verify:

• .env contains OPENROUTER_API_KEY=
• User is prompted for OpenRouter API key interactively
• No separate embedding model configuration is added

3. Interactive Mode - OpenRouter as Embedding Fallback

elizaos create test-claude-openrouter
# Select: Claude as AI model
# Select: OpenRouter as embedding model

Verify:

• .env contains ANTHROPIC_API_KEY=
• .env contains OPENROUTER_API_KEY=
• .env contains OPENROUTER_EMBEDDING_MODEL= and OPENROUTER_EMBEDDING_DIMENSIONS=

4. Non-Interactive Mode

elizaos create test-noninteractive --ai-model openrouter --non-interactive

Verify:

• .env created with placeholder: OPENROUTER_API_KEY=your_openrouter_api_key_here
• Comment included: # Get your API key from: https://openrouter.ai/keys
• Default embedding model and dimensions added

5. Regression Testing - Other Providers

# Verify other providers still work correctly
elizaos create test-openai --ai-model openai --non-interactive
elizaos create test-ollama --ai-model local --non-interactive
elizaos create test-google --ai-model google --non-interactive

Expected: All providers create projects successfully with correct .env configurations

6. Optional: E2E Testing with Real API Key

# With a real OpenRouter API key
export OPENROUTER_API_KEY=sk-or-v1-xxx...
export OPENROUTER_EMBEDDING_MODEL=openai/text-embedding-3-small
export OPENROUTER_EMBEDDING_DIMENSIONS=1536
cd packages/cli
bun run test:e2e

Discord username

neysixx