Skip to content

CodexHub 0.1.4

Choose a tag to compare

View all tags
@NOirBRight NOirBRight released this 12 Jul 05:23
· 1344 commits to main since this release
v0.1.4
d85e114
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

English

Features

  • Add Vision Proxy capability exposure for third-party models while the proxy is enabled.
  • Add an OpenAI long-context cost guard that updates Codex and Gateway together and synchronizes bound external clients.
  • Expose Official reasoning levels through Max while retaining Ultra only for supported Official models.
  • Add Stable/Beta runtime ownership isolation with explicit cross-channel takeover behavior.

Fixes

  • Preserve unified Codex conversation history across trusted provider switches and repair drift without controlling Codex processes.
  • Pool Official upstream connections, inherit the Windows proxy, cap TLS connection setup at 15 seconds, and retry only safe pre-response/zero-byte stream failures.
  • Return cancelled Official streams to the connection pool, preventing the 16-slot pool from exhausting and blocking later requests for 300 seconds.
  • Keep Official generation transport isolated from third-party Provider and OAuth refresh traffic (#60).
  • Preserve structured tool calls and tool outputs for external models, preventing completed tool calls from being repeated.
  • Support Codex Desktop title and personalized-suggestion requests using internal Responses Lite headers.
  • Keep bound external clients synchronized with context-guard changes; report applied, skipped, partial, and unconfirmed outcomes without exposing client names.
  • Replace an older same-channel managed Gateway when a newer app bundle starts, preventing a new UI from silently reusing stale backend code.
  • Show provider-scoped effective context limits in the Provider UI, including 1M for Ollama GLM-5.2 and Volc MiniMax M3.
  • Package the pinned Python/zstandard runtime required by current Codex Desktop requests.

中文

新功能

  • Vision Proxy 开启时，为第三方模型暴露视觉能力。
  • 新增长上下文费用保护，同时作用于 Codex 与 Gateway，并同步已绑定的外部客户端。
  • 官方模型开放 Light 到 Max 推理等级，Ultra 仍仅对受支持的官方模型开放。
  • 新增 Stable/Beta 运行时所有权隔离与明确的跨通道接管行为。

修复

  • 在可信 Provider 切换时保留统一的 Codex 历史对话，并在不控制 Codex 进程的前提下修复漂移。
  • 为官方上游复用连接、继承 Windows 代理、将 TLS 建连超时限制为 15 秒，并只重试响应前或零字节流等安全失败。
  • 将已取消的官方响应流归还连接池，避免 16 个连接槽耗尽后让后续请求阻塞 300 秒。
  • 隔离官方生成请求、第三方 Provider 与 OAuth 刷新请求的传输所有权（#60）。
  • 保留第三方模型的结构化工具调用与工具结果，避免已完成的工具调用被重复执行。
  • 兼容 Codex Desktop 使用内部 Responses Lite 标头发送的标题与个性化建议请求。
  • 长上下文保护变化时同步已绑定的外部客户端；区分已应用、已跳过、部分失败和无法确认，且不暴露具体客户端名称。
  • 新版应用启动时替换同通道的旧版托管 Gateway，避免新 UI 静默复用旧后端代码。
  • Provider 页面显示按 Provider 生效的上下文限制，包括 Ollama GLM-5.2 与 Volc MiniMax M3 的 1M。
  • 打包当前 Codex Desktop 请求所需的固定版本 Python/zstandard 运行时。

Issues

Closes #10
Closes #12
Closes #60

Refs #16 (generic cross-provider transport diagnosis remains post-release work)
Refs #57 (tool-protocol/capability architecture remains post-release work)

Verification

  • combined release candidate 7257701b: Python 818 passed / 1 skipped / 88 subtests; Rust 277 passed; Clippy -D warnings; UI contracts 130 passed; frontend production build passed
  • PRs #79, #80, and #81: all twelve GitHub CI jobs passed
  • stable portable candidate: CodexHub_0.1.4_portable_7257701b.zip
  • portable SHA-256: 32665404ac244724dffcfedff65cdf434706429a873b6c0d2b3b787445e3e10a
  • internal maintainer acceptance completed after testing GLM-5.2, Vision Proxy, long-context guard, external-client synchronization feedback, Gateway replacement, title requests, tool calls, and Official transport stability
  • final signed installer, updater signature, and latest.json will be rebuilt from the merged main SHA before tagging and publishing
Assets 5
Loading