Linux release assets
CLIProxyAPI_<version>_linux_<arch>.tar.gzis the default Linux build. It supports dynamic library plugins and is built against a GLIBC 2.17 baseline.CLIProxyAPI_<version>_linux_<arch>_no-plugin.tar.gzis the portable Linux build for musl-based or older systems such as OpenWrt. It does not support dynamic library plugins.
FreeBSD release assets
CLIProxyAPI_<version>_freebsd_aarch64_no-plugin.tar.gzis the FreeBSD arm64 build. It is built without CGO and does not support dynamic library plugins.
Changelog
- feat(xai): allow grok imagine aspect_ratio 9:20 (75bd6a6)
- feat(xai): also allow grok imagine aspect_ratio 20:9 (f049e00)
- fix(claude): skip retry-after header on overage-only rejections (1cce932)
- fix(codex): strip item-level and tool output prompt cache breakpoints (3662d15)
- fix(xai): restore aliased client web search tool name in responses (660a580)
- feat(codex): append (Devin) suffix to Devin model display names (2874347)
- fix(codex): normalize and support ultrafast service tier (859c486)
- feat(codex): record upstream response model and warn on silent model substitution (25f40d8)
- fix(openai): align tool call messages and preserve ordering on ambiguous outputs (cc545cb)
- feat: add PatewayAI (784285a)
- feat: add FluxA (0b9a91f)
- Merge pull request #5919 from Fesaluo/feat/xai-imagine-aspect-ratio-9-20 (05391d7)
- Merge pull request #5928 from Viggo95/feat/codex-response-model-observability (cd5af08)
- feat(executor): extend response model recording and substitution warnings to all providers (e9463ff)
- fix(executor): record authentic Devin and Gemini Interactions response models (f8467f0)
- fix(executor): robust response model observability across meta, kimi, and openai-compat streams (cde7d57)
- fix(executor): record expected Devin upstream model and bound stream observer memory (e84e248)
- fix(codex): preserve reasoning content and IDs for compat models in responses (81d6ba7)
- fix(openai): classify stream request timeout as server error (cb62a67)
- fix(translator): scope tool responses per turn to handle repeated tool call IDs (c2ea268)
- fix(executor): restrict claude advisor tool check to server tool use (f86a33f)
- test(config): verify claude cloak persistence and update behavior (f485217)
- feat(executor): support use-max-completion-tokens for openai compatibility models (690f4f3)
- fix(executor): restore hybrid passthrough mcp tool names in claude oauth (22392c5)
- fix(executor): buffer post-tool text to order devin tool calls before assistant response (b6d1f05)
- fix(schema): normalize true boolean subschemas and strip unsupported keywords (c93978c)
What's Changed
- feat(xai): allow grok imagine aspect_ratio 9:20 and 20:9 by @Fesaluo in #5919
- feat(codex): record upstream response model and warn on silent model by @Viggo95 in #5928
New Contributors
Full Changelog: v7.3.7...v7.3.8