feat: add OrcaRouter as a named embedding provider - #1252
Conversation
Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: XiaoHuo888 <sjh00112233@outlook.com>
|
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3a48c61f00
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| dimensions=dimensions, | ||
| api_key=api_key, | ||
| base_url=base_url or ORCAROUTER_DEFAULT_BASE_URL, | ||
| timeout=timeout, |
There was a problem hiding this comment.
When semantic_embedding_provider=orcarouter uses the default model, this constructor records dimensions=1536 only as Basic Memory's expected vector size, but the inherited embed_documents() call never sends a dimensions parameter to OrcaRouter. OrcaRouter's current model card for openai/text-embedding-3-small documents a 512-dimensional default unless a dimension is requested, so the normal bm reindex --embeddings path will create/expect 1536-dimensional vector storage and then fail on the first 512-dimensional response with the provider's dimension-mismatch error. Please either request dimensions=self.dimensions for OrcaRouter models that support it or fail fast unless the configured dimensions match the gateway default.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8c03fd86de
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| model_name = app_config.semantic_embedding_model or ORCAROUTER_DEFAULT_MODEL | ||
| if model_name == "bge-small-en-v1.5": | ||
| model_name = ORCAROUTER_DEFAULT_MODEL | ||
| provider = OrcaRouterEmbeddingProvider( |
There was a problem hiding this comment.
When semantic_embedding_model selects any OrcaRouter model whose native output size is not 1536 and semantic_embedding_dimensions is omitted, this branch silently constructs the provider with 1536 dimensions. Unlike the LiteLLM branch below, it does not require dimensions for custom models, so the inherited response validation fails on the first embedding request instead of rejecting the invalid configuration up front. Require an explicit dimension for non-default OrcaRouter models or resolve a model-specific default.
AGENTS.md reference: AGENTS.md:L132-L133
Useful? React with 👍 / 👎.
Summary
OrcaRouter is an OpenAI-compatible model routing gateway that brings 150+ models from OpenAI, Anthropic, Google, DeepSeek, Qwen, MiniMax and xAI behind a single endpoint and API key. Beyond routing, it runs gateway-level, zero-trust security for AI agents on the same endpoint — screening every prompt/response and governing every tool call on a default-deny basis, with no application code changes. This PR registers it as a named embedding provider so users can opt in directly.
I'm an engineer on the OrcaRouter team.
What this changes
src/basic_memory/repository/orcarouter_provider.py: newOrcaRouterEmbeddingProvider, mirroringOpenAIEmbeddingProviderbut pointed athttps://api.orcarouter.ai/v1and authenticated withORCAROUTER_API_KEY(keys start withsk-orca-).src/basic_memory/repository/embedding_provider_factory.py: registersorcarouterin both the provider factory dispatch and the provider identity resolver, withopenai/text-embedding-3-smallas the default model (1536 dimensions).src/basic_memory/config_models.py: documents theorcaroutervalue forsemantic_embedding_provider.docs/semantic-search.md: adds the OrcaRouter provider section and updates the config reference table.tests/repository/test_orcarouter_provider.py: covers lazy client construction, explicit key/base-url override, dimension mismatch, missing dependency, missing key, and factory selection/identity.Why this provider
The repo already treats
openaias a named embedding provider wired directly to a single endpoint. OrcaRouter is the same OpenAI-compatible wire, but the endpoint, key, and routing stay gateway-managed, so users get the standardprovider/modelmodel ids (e.g.openai/text-embedding-3-small) through oneORCAROUTER_API_KEY.How to use it
How this was tested
tests/repository/test_orcarouter_provider.py+ existingtest_openai_provider.py— 40 passed; fulltests/repository/suite — 605 passed, 29 skipped.ruff checkclean,ruff format --checkclean,ty checkclean on changed files.ORCAROUTER_API_KEY, the PR'dcreate_embedding_provider(provider="orcarouter")path returned 1536-dim vectors for bothembed_queryandembed_documentsthroughhttps://api.orcarouter.ai/v1/embeddings; provider identity resolves toOrcaRouterEmbeddingProvider:openai/text-embedding-3-small:1536.Related issues
None.