Problem

When using OpenRouter as an LLM provider, Letta cannot discover embedding models. The OpenRouterProvider class inherits from OpenAIProvider but doesn't implement list_embedding_models_async().

Current behavior:

• Only letta/letta-free embedding model appears in the UI/API
• OpenRouter embedding models (e.g., openai/text-embedding-3-small) are not discoverable
• Users cannot create agents with OpenRouter embeddings through normal flows

Root cause:
OpenRouter's /models endpoint only returns LLM models, not embedding models. Unlike OpenAI's API, there's no /models response that includes embedding model metadata.

Environment

• Letta version: 0.15.1 (Docker)
• Provider: OpenRouter with OPENROUTER_API_KEY set

Proposed Solution

Add list_embedding_models_async() method to OpenRouterProvider class in letta/schemas/providers/openrouter.py.

Since OpenRouter doesn't expose embedding models via their API, the solution is to return hardcoded EmbeddingConfig objects for the known supported models:

async def list_embedding_models_async(self) -> list[EmbeddingConfig]:
    """
    Return hardcoded OpenRouter embedding models.
    OpenRouter's /models endpoint doesn't list embedding models,
    but they support OpenAI's embedding models via the /embeddings endpoint.
    """
    return [
        EmbeddingConfig(
            embedding_model="openai/text-embedding-3-small",
            embedding_endpoint_type="openai",
            embedding_endpoint=self.base_url,
            embedding_dim=1536,
            embedding_chunk_size=DEFAULT_EMBEDDING_CHUNK_SIZE,
            handle=self.get_handle("openai/text-embedding-3-small", is_embedding=True),
            batch_size=DEFAULT_EMBEDDING_BATCH_SIZE,
        ),
        EmbeddingConfig(
            embedding_model="openai/text-embedding-3-large",
            embedding_endpoint_type="openai",
            embedding_endpoint=self.base_url,
            embedding_dim=3072,
            embedding_chunk_size=DEFAULT_EMBEDDING_CHUNK_SIZE,
            handle=self.get_handle("openai/text-embedding-3-large", is_embedding=True),
            batch_size=DEFAULT_EMBEDDING_BATCH_SIZE,
        ),
    ]

Testing Done

Tested with a patched openrouter.py mounted via Docker volume:

Additional Context

• OpenRouter uses the same /embeddings endpoint format as OpenAI
• The embedding_endpoint_type="openai" ensures compatibility
• This is a minimal change that unblocks OpenRouter users
• Happy to submit a PR if this approach is acceptable

Related Issues

• OpenRouter API support #1670 - OpenRouter API support (closed, general support added)
• Getting error when calling archival_memory_insert #1784 - archival_memory_insert error with OpenRouter (closed, was about API key config)