Releases: zzccchen/token-pulse
Release list
TokenPulse 0.1.1 — Windows portable preview
TokenPulse 0.1.1 adds a Windows portable download. Extract the entire ZIP and open TokenPulse.exe; keep the _internal folder beside it. Python does not need to be installed separately.
Downloads
- Windows x64:
TokenPulse-0.1.1-windows-amd64.zip(about 93 MiB). - Python wheel and source archive for source/package installations.
SHA256SUMS.txtcovers all three files.
The portable ZIP includes external, replaceable Qt DLLs; LGPLv3/GPLv3 texts and upstream attribution notices; complete matching Qt Base and PySide/Shiboken source archives; the corresponding TokenPulse source snapshot; and library replacement/rebuild instructions. See START-HERE.txt and notices/WINDOWS-DISTRIBUTION.md after extraction. TokenPulse's own code remains MIT-licensed.
Changes
- Add a portable directory build with runtime metadata and checksums.
- Include reviewed license/source materials and remove unused Qt plugins and the software OpenGL renderer.
- Update English/Chinese introductions and list Codex CLI compatibility as 0.153.4–0.160.0.
Validation and limits
218 offline tests, Ruff, formatting, documentation links and source/wheel builds passed from a clean checkout. The wheel was installed and smoke-tested outside the source tree. The final portable ZIP passed file/source hashes, archive checks and isolated native demo startup/exit checks for normal-window, tray and hidden-start modes. A process-module check confirmed external Qt DLL loading.
The Windows build is unsigned. Clean-machine Windows testing and broader native Linux desktop tray testing remain pending. There is no Linux binary in this release. Speeds remain completed-output estimates from local logs, not controlled benchmarks.
Built with Codex. If you run into a problem, please open an issue. Independent project, not affiliated with or endorsed by OpenAI.
TokenPulse 0.1.0 — First public preview
TokenPulse 0.1.0 — First public preview
TokenPulse (词脉) is a local, read-only Codex tray monitor for Windows and Linux. See completed-output speed across conversations, inspect evidence behind mode labels, and keep history on your own machine.
Included
- Per-model weighted TPS, median, range, and scatter plots over 30 minutes, 1 day, and 7 days.
- Local pseudonymized history, filters, and CSV export.
- Separate configuration, submitted-setting, request, and actual-tier evidence.
- Tray and normal-window entry points, light/dark themes, reduced effects, and English/Chinese UI.
- Synthetic demo mode, bilingual onboarding, and contributor documentation.
Try the demo from source
Install Python 3.11+ and uv, then run:
git clone --branch v0.1.0 https://github.com/zzccchen/token-pulse.git
cd token-pulse
uv sync --locked
uv run token-pulse --demoThe demo needs no Codex account. Close it and run uv run token-pulse to observe existing local sessions.
The attached wheel is also available for an isolated Python tool installation after downloading it:
uv tool install --python 3.13 ./token_pulse-0.1.0-py3-none-any.whl
token-pulse --demoThe wheel needs Python and installs Qt as a dependency. SHA256SUMS covers the attached wheel and source archive, not GitHub's automatically generated source archives.
Validation and boundaries
- Windows / Ubuntu, Python 3.11 / 3.13: all four GitHub Actions jobs passed tests, lint, format, local documentation links, demo startup/exit, and package builds. Run
- Local Windows: 205 offline tests and isolated installed-wheel demos in English and Chinese passed.
- Public snapshot and original development history checked with Gitleaks; no matches found. Package contents and project/dependency notices reviewed.
- Native Windows tray behavior was tested during development. WSL Wayland validation does not establish native GNOME/KDE/Xfce tray compatibility.
This is an early preview targeting observed Codex 0.153.4-series internal logs. Missing boundaries produce no TPS; estimates are not a controlled benchmark or a live token counter. Final server-confirmed tiers are not currently extracted. No Windows EXE, AppImage, or PyPI release is included.
English guide · 中文说明 · Known limits
词脉首个公开预览版:支持本机 Codex 按模型测速、散点时间轴、本地历史与 CSV、中英文及深浅色界面。建议先运行隔离演示;测速是基于日志的已完成输出估算,完整 Linux 桌面托盘仍待验证。欢迎使用合成数据报告问题。
Independent community project, not affiliated with or endorsed by OpenAI.
History reconstruction (2026-09-22)
The original single-commit public baseline has been replaced with 17 dependency-ordered development commits. Each stage passes its available tests, lint/format checks, and package builds. These commits document a verified reconstruction, not the original development timestamps.
The v0.1.0 tag now points to 727359d8e0b36d89e94f82a9b3c503bb98a2216f. Its final file tree is identical to the original preview; the attached wheel, source archive, and checksums are unchanged. Existing clones should preserve any local work before switching to the reconstructed history; a fresh clone is the simplest option if no local work needs retaining.
The reconstructed main branch, preview tag, and rebased dependency PR each passed the Windows/Ubuntu × Python 3.11/3.13 CI matrix. Main checks · Tag checks.