It’s getting a little painful
I won’t say much about latency and stuff. I talked about it yesterday.
Today, it seems all I did was make this usage visualization (half-finished; the refresh function is currently broken, so please wait for 0.6.2).
But it’s so ugly…
Still, putting off the release for another day wouldn’t really be right.
All the features work, but it’s just too ugly.
Currently giving it a major overhaul.
If you have any suggestions, please share them. Thanks, meow. Release v0.6.1 · HEXUXIU/M365-Copilot2API
Let’s include the speed metrics as well.
metric min med max mean
out_tokens 662 959 1131 924
ttfb (s) 0.02 0.03 0.17 0.05
total (s) 25.75 34.25 40.67 33.84
tps (tok/s) 23.6 26.9 33.2 27.4
Could you add a hot update feature? Otherwise, downloading the binary, uploading it, and setting permissions is way too troublesome—and the process is super laggy, too.
Updating this kind of binary is pretty convenient—you can just write a script to fetch the GitHub version number, compare the hash value, and update it periodically or manually.
I did this on purpose…
I don’t like load balancing
Because that means I’d have to warm up all the accounts, and I don’t want to waste my attempts
I rely on failover—only switching to another account when one is used up
│ >_ OpenAI Codex (v0.150.1) │ │ │ │ model: gpt-5.6-sol /model to change │ │ directory: C:\WINDOWS\system32 │ ╰───────────────────────────────────────────╯ Tip: New Build faster with Codex. › hi ◦ Reconnecting… 4/5 (8s • esc to interrupt) └ Unexpected status 401 Unauthorized: Incorrect API key provided: m365****. You can find your API key at https:// platform.openai.com/account/api-keys., url: wss://api.openai.com/v1/responses, cf-ray: a319eb750e0c58c9-MAA › Ask Codex to do anything
I am just letting you know that there is such a help page for using Codex on the web. I am not saying that what is displayed on it is necessarily correct.