v1.1.0

Behaviour change

max_tokens, max_completion_tokens and max_output_tokens are now enforced, at about four characters per token. max_completion_tokens and max_output_tokens also count reasoning, as in OpenAI's API, so a small limit on a reasoning model can return a short or empty answer with finish reason length (incomplete in Responses). Clients that send small limits by habit should raise them or leave them out.

Install

npx zengate@1.1.0
# or
docker run -d --name zengate -p 127.0.0.1:8083:8083 -v zengate:/data developingchet/zengate:1.1.0

Image digest: sha256:e031d1b70210d9eb74628daefd4283984b0ab39de82cb25aa92f263c769a097e

Verify the image signature

cosign verify developingchet/zengate:1.1.0 \
  --certificate-identity-regexp '^https://github.com/developingchet/zengate/.github/workflows/release.yml@refs/tags/v' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com

Verify the release files

checksums.txt and the package tarball are signed with cosign (*.sigstore.json):

cosign verify-blob checksums.txt --bundle checksums.txt.sigstore.json \
  --certificate-identity-regexp '^https://github.com/developingchet/zengate/.github/workflows/release.yml@refs/tags/v' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com
sha256sum -c checksums.txt

The npm package carries provenance: npm audit signatures.

What's Changed

• ci: stop running the Socket scan weekly by @developingchet in #12
• ci: run the dependency audit as its own Vulnerability Check job by @developingchet in #13
• fix: wait for a longer stop sequence that starts before a shorter match by @developingchet in #14
• fix: keep OpenCode turns alive after rejected tools by @developingchet in #15
• fix: harden the gateway and add max_tokens, legacy completions and Prometheus metrics by @developingchet in #16
• chore: release v1.1.0 by @developingchet in #17

Full Changelog: v1.0.4...v1.1.0