Openai Compatible Embeddings Batching

Custom embeddings adapters against OpenAI-compatible endpoints (OpenRouter, vLLM, Together, Anyscale, Groq, self-hosted gateways) must batch transparently and fail loudly — some providers return HTTP 200 with a non-standard error body missing the `data` key, which surfaces as a bare KeyError deep inside LangChain. Includes the two-layer adapter fix and the regression test that actually exercises the batching path.

aksheyw Updated
How most of these skills got written. A bug eats more than 20 minutes of a working session. At session end I write down the symptoms, the root cause, the exact

File contents

aksheyw/claude-code-learned-skills commit 5e2028e7f3

Frequently asked questions

npx get-skill add aksheyw/openai-compatible-embeddings-batching