<!-- library-gap-audit: fastembed-py -->
Gap
fastembed (Qdrant's ONNX-based embeddings library) has no Braintrust instrumentation in this repo. It exposes a distinct top-level execution API — TextEmbedding, SparseTextEmbedding, ImageEmbedding, and LateInteractionTextEmbedding classes, each with an .embed() / .query_embed() / .rerank() method that runs local ONNX inference — and is one of the most widely used embeddings-execution libraries in the Python generative-AI ecosystem, especially in RAG pipelines.
This is distinct from the embeddings surfaces already instrumented in this repo (openai, cohere, google_genai, mistral, huggingface_hub, transformers) — fastembed has its own independent client/model API and does not route through any of those SDKs.
Weekly downloads: 1,514,406 (as of 2026-08-17; source: https://pypistats.org/api/packages/fastembed/recent)
Braintrust docs status: not_found — checked https://www.braintrust.dev/docs/integrations; FastEmbed/Qdrant embeddings are not listed among the supported integrations or SDK/provider grids. The page's only guidance for unlisted providers is to use the generic AI Proxy or request an integration.
Upstream sources:
Local repo files inspected (confirming no existing instrumentation):
py/src/braintrust/integrations/ (recursive listing) — no fastembed directory among adk, agentscope, agno, ai_sdk, anthropic, autogen, bedrock_runtime, claude_agent_sdk, cohere, crewai, cursor_sdk, dspy, google_genai, huggingface_hub, instructor, langchain, litellm, livekit_agents, llamaindex, mistral, openai, openai_agents, openrouter, pipecat, pydantic_ai, strands, temporal, transformerspy/src/braintrust/wrappers/*.py — no fastembed wrapperpy/pyproject.toml [tool.braintrust.matrix] and [tool.braintrust.vendor-packages] — no fastembed entrypy/noxfile.py — no fastembed test session- Repo-wide case-insensitive grep for
fastembed — no matches outside this issue's own research
Suggested instrumentation: wrap TextEmbedding.embed() / .query_embed() (and the sparse/image/late-interaction variants) similarly to the existing embeddings-execution wrappers, capturing model name, input count/size, and output vectors/metadata as span data.
<!-- library-gap-audit: fastembed-py -->Gap
fastembed(Qdrant's ONNX-based embeddings library) has no Braintrust instrumentation in this repo. It exposes a distinct top-level execution API —TextEmbedding,SparseTextEmbedding,ImageEmbedding, andLateInteractionTextEmbeddingclasses, each with an.embed()/.query_embed()/.rerank()method that runs local ONNX inference — and is one of the most widely used embeddings-execution libraries in the Python generative-AI ecosystem, especially in RAG pipelines.This is distinct from the embeddings surfaces already instrumented in this repo (
openai,cohere,google_genai,mistral,huggingface_hub,transformers) —fastembedhas its own independent client/model API and does not route through any of those SDKs.Weekly downloads: 1,514,406 (as of 2026-08-17; source: https://pypistats.org/api/packages/fastembed/recent)
Braintrust docs status:
not_found— checked https://www.braintrust.dev/docs/integrations; FastEmbed/Qdrant embeddings are not listed among the supported integrations or SDK/provider grids. The page's only guidance for unlisted providers is to use the generic AI Proxy or request an integration.Upstream sources:
0.8.0(2026-03-23)Local repo files inspected (confirming no existing instrumentation):
py/src/braintrust/integrations/(recursive listing) — nofastembeddirectory among adk, agentscope, agno, ai_sdk, anthropic, autogen, bedrock_runtime, claude_agent_sdk, cohere, crewai, cursor_sdk, dspy, google_genai, huggingface_hub, instructor, langchain, litellm, livekit_agents, llamaindex, mistral, openai, openai_agents, openrouter, pipecat, pydantic_ai, strands, temporal, transformerspy/src/braintrust/wrappers/*.py— no fastembed wrapperpy/pyproject.toml[tool.braintrust.matrix]and[tool.braintrust.vendor-packages]— nofastembedentrypy/noxfile.py— nofastembedtest sessionfastembed— no matches outside this issue's own researchSuggested instrumentation: wrap
TextEmbedding.embed()/.query_embed()(and the sparse/image/late-interaction variants) similarly to the existing embeddings-execution wrappers, capturing model name, input count/size, and output vectors/metadata as span data.