Usage and context
Saiki shows how much of your Claude and ChatGPT limits you've used and when they reset, and how full each chat's context is, so you can pace yourself.
AI Usage
Type usage (or "limits", "quota") and open AI Usage. It shows your subscriptions' limits, Claude first, then ChatGPT. Press ↵ to refresh, and esc to go back.
For each AI you see:
- your plan, and Limit reached when you've hit one;
- when it was last updated;
- the tool's version, and whether a newer one is out;
- the account's email, when the tool reports one;
- one bar per limit window: 5-hour, Weekly, a weekly limit for one model ("Weekly · …"), or Monthly, whichever your plan has.
Reading a limit
| Part | What it tells you |
|---|---|
| "34 % used" | How much of this window you've used. It turns orange from 75 % and red from 90 %. |
| The thin mark on the bar | How much of the window's time has passed. If the bar is past the mark, you're using it faster than evenly. |
| "Resets in 2h 10m" | When the window starts over. Just after, it says "Just reset". |
| "At this pace, …" | A forecast from your use so far: when it would run out, or that it lasts until it resets, and about how much you'll have used by then. |
An account signed in with an API key has no subscription limits, and AI Usage says so. If a tool isn't installed, its section says that instead.
The meter in a chat
At the right of a chat's header there's a small bar. It's filled to whichever is fuller: the chat's context or your tightest limit. It stays quiet until 50 %, then shows the number beside it, orange from 75 % and red from 90 %. Point at it to see both numbers.
Click it (or type /status) to open both meters:
- Context: how full it is, in percent and tokens, with Compact Now.
- This AI's usage: each window, how much is used and when it resets, and AI Usage… for the full screen.
Context and Compact Now
Every chat has a context window: how much of the conversation the AI can keep in mind at once. 100 % is the point where the tool would compact the chat on its own.
- Above 80 %, a line under the header says how full it is, with Compact Now.
- Compact Now, or typing
/compact, has the AI summarize the chat so far to free room. The chat carries on from the summary. Compacting uses some of your limits.
Compact first?
Sending a message to a chat that holds 100,000 tokens or more and has been idle for 70 minutes makes the AI read the whole conversation again, which uses a lot of your 5-hour limit. So Saiki holds your message and asks Compact first?, saying how long it's been idle and how many tokens it holds:
- ⌘↵ Compact, Then Send: compacts, then sends your message.
- ↵ Send Anyway.
- esc puts your message back in the field.
Usage alerts
Saiki tells you with a notification when a window passes 80 % ("Claude: 80 % of your 5-hour limit used") and when you reach a limit ("Claude limit reached"), each with when it resets. Each is sent once per window until it resets. Open shows AI Usage.
Where the numbers come from
The numbers are the ones Claude Code and Codex report about your own account, checked when a chat starts and after answers. Saiki never reads your sign-in to get them. When no Claude chat is running, AI Usage starts Claude Code briefly with everything else off (no tools, no saved session) just to ask for your limits.