Claude knows its limits

Multee

Rename the config keys

Tests for parse_args

Which config keys does the app read, and where?

Read(config.py)

Read 42 linesshow

Six keys, all in camelCase: apiKey, baseUrl, units, timeout, cacheDir and retryCount. config.py loads them, and they're read all over the code.

Rename the config keys to snake_case everywhere, and update the docs.

Grep("apiKey|baseUrl" (*.py))

Found 7 files

Update(config.py)

6 additions, 6 removals

Update(cli.py)

3 additions, 3 removals

multee · usage_status

Context (this conversation): 82,140 of 200,000 tokens used (41%), 117,860 left.

5-hour limit: 86% used, resets in 1h 12m (Sat 10 Oct, 19:50 GMT+5:30). At the pace of the last half hour it is full in about 48m — before it resets.

7-day limit: 61% used, resets in 3d 4h (Tue 13 Oct, 22:50 GMT+5:30).

(Limits as last reported to one of Multee's chat tabs; with only terminal tabs working they can be behind.)

… +10 lines  show more

multee · usage_plan(sonnet)

Noted — the other sessions see it in usage_status until this turn ends.

Switching to Sonnet 5.5 — from your next step. The user sees a line in the chat saying so, and Multee puts it back as it was once the limits have reset (unless the user changes it first). Mention it in your reply.

Claude switched to Sonnet 5.5 — 5h at 86%; back at 19:50.

Update(docs/config.md)

8 additions, 8 removals

Renamed the keys in 7 files and the docs. I moved myself to Sonnet 5.5 for the mechanical part: the 5-hour limit is at 86%, and “Tests for parse_args” is using it too. Multee puts Opus back after the reset.

The usage limit has reset — back to Opus 5.5, as before Claude switched down for it.

Now update the README to match.

✻Crafting…· 14m 2s

✻Working…· 14m 2s

✻Pondering…· 3s

Message Claude…  / commands · @ files & agents · ! shell

⏵⏵ accept edits on (⇧⇥)ucO 5.5 ▾S 5.5 ▾O 5.5 ▾5h1h1m4h7d3d

main

docs

tests

cli.py

config.py

README.md

weather.py

SESSIONS

multee

weather-cli

1. 1Reach 80% The 5-hour limit passes 80%, and Multee warns every working Claude at its next step.
2. 2Check the numbers Claude reads usage_status: its context, both limits, when the 5-hour one fills, and who else is working.
3. 3Step down a model Claude moves itself to Sonnet; Multee says so in the chat, and the status line shows S 5.5.
4. 4Switch back at the reset When the limit resets, Multee puts Opus back by itself and says so; your next message runs on Opus.

Good to know

Rename the config keys to snake_case everywhere, and update the docs.

Grep("apiKey|baseUrl" (*.py))

Found 7 files

Update(config.py)

6 additions, 6 removals

Update(cli.py)

3 additions, 3 removals

multee · usage_plan(low)

Effort set to low — from your next step. The user sees a line in the chat saying so, and Multee puts it back as it was once the limits have reset (unless the user changes it first). Mention it in your reply.

Claude lowered its effort to low — 5h at 92%; back at 19:40.

Message Claude…  / commands · @ files & agents · ! shell

⏵⏵ accept edits on (⇧⇥)ucO 5.5 ▾5h42m7d3d

Effort: low — how hard Claude thinks before answering. Drag or click to change.

Lower effort instead of a cheaper model

Instead of a cheaper model, Claude can turn down its effort: how hard it thinks before answering. The chat says so, the effort slider on the status line moves, and Multee puts it back after the reset too.

Run the tests and fix whatever fails.

Read(tests/test_weather.py)

Read 64 linesshow

Bash(python3 -m pytest tests/ -q)

2 failed, 12 passed

multee · usage_plan(opus)

No switch to Opus 5.5: it isn't cheaper than the model you are on (Sonnet 5.5). usage_plan only moves a conversation down; a stronger model is the user's to pick.

Staying on Sonnet 5.5. Switch to Opus from the model menu if these need it.

Message Claude…  / commands · @ files & agents · ! shell

⏵⏵ accept edits on (⇧⇥)ucS 5.5 ▾5h1h7d3d

Only ever down, never up

Claude can switch only to a cheaper model or a lower effort, and only while a limit is at 80% or more. Asked for more, it’s refused with the reason: a stronger model is yours to pick. Only chat tabs switch; terminal tabs just get the numbers.

Claude

USAGE LIMITS

Continue terminal sessions when a limit resets

A Claude terminal tab stopped at the 5-hour or weekly limit is told to carry on. A tab you typed in meanwhile is left alone.

Let Claude lower its own model for limits

When a limit is at 80% or more, Claude in a chat tab can move itself to a cheaper model or lower effort. The chat says so, and Multee puts it back when the limit resets. Never to a stronger model.

Turn the switching off

It’s on by default. Switch off Let Claude lower its own model for limits in Settings ▸ Claude, and Claude still sees the numbers and the warning but never changes its own model or effort.

Related

Usage limits The 5h / 7d meters→ Two bars in a chat's status line show how full Claude's 5-hour and 7-day usage limits are. Usage limits Carrying on after a limit→ A chat stopped at Claude's usage limit carries on by itself once the limit resets. Usage limits Everything in Usage limits→ Every change to Usage limits, newest first.