<!-- library-gap-audit: portkey-ai-py -->
Gap
portkey-ai (the Python SDK for the Portkey AI Gateway) has no Braintrust instrumentation in this repo. It provides a unified, OpenAI-SDK-compatible execution surface (Portkey(...).chat.completions.create(...), .completions.create(...), .embeddings.create(...), streaming, and tool-calling) that routes requests to 250+ upstream LLM providers via virtual keys, fallbacks, load balancing, and semantic caching.
This is directly analogous to the gateway/router libraries already instrumented in this repo — litellm and openrouter — but portkey-ai itself has no wrapper, despite having comparable or higher adoption and the same execution-oriented surface (generation, streaming, tool calls) that the existing gateway integrations cover.
Weekly downloads: 830,380 (as of 2026-08-17; source: https://pypistats.org/api/packages/portkey-ai/recent)
Braintrust docs status: not_found — checked https://www.braintrust.dev/docs/integrations; Portkey is not listed among the supported integrations, providers, or gateway/proxy tooling. 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 portkey directory (contrast with existing gateway coverage: litellm, openrouter)py/src/braintrust/wrappers/*.py — no portkey wrapperpy/pyproject.toml [tool.braintrust.matrix] and [tool.braintrust.vendor-packages] — no portkey/portkey-ai entrypy/noxfile.py — no portkey test sessionpy/src/braintrust/_generated_types.py — only false-positive substring match (...ReportKey...), not a real reference- Repo-wide case-insensitive grep for
portkey — no genuine matches
Suggested instrumentation: wrap the Portkey client's chat.completions.create, completions.create, and embeddings.create methods (including streaming) similarly to the existing litellm/openrouter gateway wrappers, capturing the routed provider/model, request/response payloads, and token usage as span data.
<!-- library-gap-audit: portkey-ai-py -->Gap
portkey-ai(the Python SDK for the Portkey AI Gateway) has no Braintrust instrumentation in this repo. It provides a unified, OpenAI-SDK-compatible execution surface (Portkey(...).chat.completions.create(...),.completions.create(...),.embeddings.create(...), streaming, and tool-calling) that routes requests to 250+ upstream LLM providers via virtual keys, fallbacks, load balancing, and semantic caching.This is directly analogous to the gateway/router libraries already instrumented in this repo —
litellmandopenrouter— butportkey-aiitself has no wrapper, despite having comparable or higher adoption and the same execution-oriented surface (generation, streaming, tool calls) that the existing gateway integrations cover.Weekly downloads: 830,380 (as of 2026-08-17; source: https://pypistats.org/api/packages/portkey-ai/recent)
Braintrust docs status:
not_found— checked https://www.braintrust.dev/docs/integrations; Portkey is not listed among the supported integrations, providers, or gateway/proxy tooling. 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) — noportkeydirectory (contrast with existing gateway coverage:litellm,openrouter)py/src/braintrust/wrappers/*.py— no portkey wrapperpy/pyproject.toml[tool.braintrust.matrix]and[tool.braintrust.vendor-packages]— noportkey/portkey-aientrypy/noxfile.py— noportkeytest sessionpy/src/braintrust/_generated_types.py— only false-positive substring match (...ReportKey...), not a real referenceportkey— no genuine matchesSuggested instrumentation: wrap the
Portkeyclient'schat.completions.create,completions.create, andembeddings.createmethods (including streaming) similarly to the existinglitellm/openroutergateway wrappers, capturing the routed provider/model, request/response payloads, and token usage as span data.