v0.1.2

Highlights

• Uses the client-facing RequestedModel for Gemini scoping and alias restoration.
• Reassembles fragmented raw JSON and SSE callbacks with bounded per-request state.
• Emits complete SSE data: JSON lines whether or not they include a trailing newline.
• Preserves already-compliant token usage without double counting reasoning tokens.
• Builds Linux amd64 and arm64 artifacts against a verified GLIBC 2.17-or-older baseline.

Verification

• go test -race ./... and go vet ./...
• Seven platform archives and SHA-256 checksums verified
• Production CPA and LinuxDo Hub regression matrix passed for four Gemini models across Chat, Responses, and both streaming APIs

v0.1.1

Superseded by v0.1.2. Do not deploy v0.1.1: it can withhold complete Responses API SSE data lines that arrive without a trailing newline, causing streaming requests to end with HTTP 502.

v0.1.0

Automated release for v0.1.0.