v6.9.43

Changelog

• b299429 fix(responses): align codex responses events and routing

v6.9.42

Changelog

• 7e175af fix(openai-compat): strip metadata for unsupported upstreams

v6.9.41

Changelog

• e50d4c8 chore(release): set default version to v6.9.41
• dff6071 optimize proxy transport reuse and reduce OpenAI SSE write overhead

v6.9.40

Changelog

• cbec7eb fix(api): enrich healthz and simplify server bootstrap
• 878a7d3 fix(logging): keep api response error formatting stable
• 20cd6ce fix(logging): reduce request log memory churn
• fee6c4d fix(openai): preserve compat passthrough fields
• 3225129 fix(responses): repair openai responses conversion

v6.9.39

Changelog

• fe25e46 fix(auth): persist runtime scheduler state across restarts

v6.9.38

Changelog

• e61ebc2 fix(handlers): merge provider matches for suffixed models
• 403aff8 fix(usage): cap retained request details per model
• ed226de fix(usage): persist statistics snapshots across restarts

v6.9.37

Changelog

• 07b0f84 chore(git): add line ending normalization rules
• 61a6618 feat(routing): add sticky round robin routing

v6.9.36

Changelog

• 27fbc83 refactor(module): migrate self imports to fork path

v6.9.35

Changelog

• 90a2442 fix(usage): reset first-byte latency per upstream attempt

v6.9.34

Changelog

• be3aa8a fix(release): use tracked embedded models catalog