Skip to content

v1.1.10

Choose a tag to compare

View all tags
@github-actions github-actions released this 03 Jul 08:35
· 38 commits to main since this release
v1.1.10
This tag was signed with the committer’s verified signature.
MitulShah1 Mitul Shah
GPG key ID: 9A023A4CFFD1D25D
Verified
Learn about vigilant mode.
1104f9a
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

[1.1.10] — 2026-07-03

Proxy & governance hardening release. Corrects the /v1/* pass-through proxy so it no longer doubles the /v1 path segment and no longer forwards OpenAI-shaped requests to providers whose upstream API is not OpenAI-compatible, and extends per-key budget governance to the embeddings and image-generation endpoints. Adds an optional provider request-signing hook and a shared base-URL validator. No breaking API changes relative to v1.1.9 — the Provider, ProxiableProvider, and plugin contracts are unchanged.

Fixed

  • Pass-through proxy path doubling: the /v1/* pass-through no longer duplicates the /v1 segment for providers whose configured base URL already ends in /v1 (e.g. https://api.x.ai/v1), which previously produced a /v1/v1/... upstream path and a 404. Both base-URL conventions now forward correctly.

Changed

  • Pass-through proxy scope (operator-visible): the /v1/* pass-through now returns HTTP 501 for providers whose upstream API is not OpenAI-wire-compatible — Anthropic, Google Gemini, AWS Bedrock, Cohere, Vertex AI, and Azure OpenAI/Foundry — instead of forwarding an OpenAI-shaped request their API cannot process (which failed upstream, and for AWS Bedrock under SigV4 was sent unauthenticated). These providers remain fully available through their translated chat/completions, embeddings, and images/generations endpoints.
  • Per-key budget on embeddings and images (operator-visible): per-key budget limits now apply to /v1/embeddings and /v1/images/generations, not only chat. Embedding requests are gated and their spend recorded; image generation is gated (it reports no token usage). Per-IP rate limiting, authentication, and the request body-size limit already applied to every endpoint.

Added

  • Provider request-signing hook: an optional RequestSigner provider interface lets a provider sign each outbound pass-through request (e.g. AWS SigV4); the proxy invokes it when a provider implements it. No in-tree provider signs yet — this is the seam for future signed pass-through.
  • Shared base-URL validation: a shared core.ValidateBaseURL helper validates that a provider base URL is an http(s) URL with a host, adopted by the Anthropic provider. Remaining providers migrate to it in subsequent releases.

Contributors

Thanks to everyone who shipped this release:

Full changelog: https://github.com/ferro-labs/ai-gateway/blob/v1.1.10/CHANGELOG.md

Contributors

MitulShah1
Assets 8
Loading