feat: add usage / quota visibility for Codex sessions and accounts

[Fanzzzd]

I’d like T3 Code to surface usage information so users can better understand cost/rate-limit consumption while working.

Why

For long-running coding sessions it is hard to tell:

• how much usage a thread consumed
• whether the current account is close to rate limits
• whether a model/service-tier choice is still appropriate

Requested scope

• Show per-thread or per-turn token usage when available.
• Show account-level rate limit / credits / quota information when available from the provider.
• Add a lightweight usage summary in the UI, ideally without overwhelming the chat view.

Suggested UX

• Thread-level token usage near the thread/session metadata
• Account usage summary in sidebar or settings/account area
• Refresh or auto-update on relevant provider events

Suggested acceptance criteria

• Users can inspect current thread/session usage without opening logs.
• Users can inspect account-level quota or rate-limit status when the provider exposes it.
• The UI degrades gracefully when the provider does not expose usage data.

Relevant code context

• apps/server/src/codexAppServerManager.ts
• apps/server/src/provider/Layers/CodexAdapter.ts
• apps/web/src/components/ChatView.tsx

Reference implementation ideas

• CodexMonitor exposes local usage snapshots, thread token usage, and account rate-limit views.
• Codex app-server events include account and token-usage style notifications that look relevant to this problem.

[maria-rcks]

Not planned at this time. If yall have any examples of similar open sources apps doing it well, share them here.

[nassimna]

This is important, and every app should have this since it's important to switch threads every time you reach the window context limit, because most of the models lose accuracy after compact
the best example is codex and every CLI

[maria-rcks]

@nassimna im mainly talking about usage visibility, sorry for not being clear on my previous message

[Sayrix]

A pie chart, similar to what GitHub Copilot uses, might fit well with the current UI of T3 Code.

[Aditya190803]

@Fanzzzd @maria-rcks

I’ve implemented this for Codex and I’m planning to open a PR. It looks like an L-XL sized change, and I’ve attached screenshots for context.

I should mention I’m currently unvouched, so please let me know if that affects whether you’d like me to open the PR.

This is the settings panel:

This is the ContextWindowMeter:

[juliusmarminge]

I don't think we need to expose it in the chat view, it just feels awkward to have it when you hover a gauge which shows 35 to see aa completely unrelated percentage for a completely unrelated metric.

but i do think we could expose it under the provider view. UI there could probably use a little work to expose both the session and weekly limits. and we should do it for claude as well!

if ur up for it i can review the pr

[Aditya190803]

Sure, I can take that on.

Could you clarify what you mean by session limits?
Are you referring to the context window or something else.
@juliusmarminge

[juliusmarminge]

both codex and claude have 2 buckets, one burst (often referred to as session limit) and one weekly

[Aditya190803]

Hey @juliusmarminge — I’ve opened a PR for this here:
#1605

Could you check it out when you get a chance?

[sethigeet]

Why not show it in the model picker, similar to how the Codex App does it?

[LukasKH]

What is the status on this?

[Aditya190803]

The PR is open and ready to be merged
It just needs approval from Julius. He will review it after returning from Miami.

[jonocodes]

Looking forward to this, as I use this feature in claude code web quite a bit.

Note the fields seem to change quite a bit as features change. For example, today there is a weekly Fable allowance on my subscription. But next week that may not exist.

[NicL9923]

+1 on wanting this, and some additional user perspective:

On Claude Max ($100) and GPT Plus ($20), I have a pretty good feel for how much I can get done, but I still end up checking pretty frequently, especially as I get closer to the limit. I can imagine the GPT $100+ and Claude Max 5x plans hardly ever feeling the need to check this, but here on these lower plans I definitely feel the need to keep a closer eye on it. In Settings->Providers would be better than nothing, but ideally it's visible in the chat interface, and I kinda dig the context window one Aditya showed ... perhaps a setting?

Related - this is especially desirable as an extra data point as currently the sessions kinda just stay in limbo with no indication as to why when usage runs out

@Aditya190803 glad to help if you need anything

[NicL9923]

@juliusmarminge If I may be so bold ^. Will be keeping a close eye to respond quickly to any feedback/etc provided y'all don't have any blocking objections. Scoped to Claude/Codex only as those are more straightforward and I'd have to guess the main two providers used

[haseeb-heaven]

Anyone working one this? Much needed feature

[elijahbutler]

What do you think!

[jonocodes]

May consider using Codexbar cli for this
https://github.com/steipete/CodexBar/blob/main/docs/cli.md

I have been using it a bit as it seems to be a good abstraction to fetch quotas across providers.

[juliusmarminge]

Related proposals being consolidated here: #6896 specifies remaining limits, reset times, manual refresh and stale-data indicators; #8148 links a fork implementation with per-account Codex/Claude quota handling. These are live subscription limits, separate from the existing token/cost Usage page.

[mikenrafter]

I have a fork I made and use that displays this data for Claude and Cursor subscriptions:
https://github.com/mikenrafter/t3code/tree/feat/live-quota-rebased

and I use the web version over tailscale on mobile because I can't live without this feature (and the mobile app, last I checked doesn't have published source code)

https://github.com/mikenrafter/t3code/tree/feat/mobile-panel-fullscreen

^ includes some mobile-friendly web changes in addition to the quota implementation.

I'm sure this could be done in a better manner, but this works for me and is good fodder for a more full solution.