For Claude Code on Pro and Max plans
Hit the limit.
Keep the prompt.
When Claude Code says you've hit your limit, the prompt you just typed is gone. hi-ai catches it, waits for the reset, and sends it into the same session for you. It also says hi the moment each five-hour window ends, so the next one starts on time instead of when you get back.
$ npm install -g hi-ai && hi-ai setupmacOS or Linux, Node 22+, Claude Code signed in. Then keep working as usual. What setup changes
Watch one afternoon
Left: your Claude Code session. Right: the hi-ai worker in the background. The output is what the real tool prints.
Your five-hour windows, with and without hi-ai
Claude's five-hour window starts with your first prompt, not on a clock. Set your day and see where the windows land.
Assumes your Mac is awake at the reset, and that a window you start is a window you use. hi-ai never raises your limit. It stops windows from sitting idle.
How it works
Two Claude Code hooks and one small background worker. No server, no account.
-
1
Catch
A
UserPromptSubmithook remembers your last prompt in each session. When the limit stops a turn, theStopFailurehook queues that prompt, with the session, folder and permission mode it came from. -
2
Wait
The worker reads the exact reset time from Claude Code's own
rate_limit_event. No guessing, no polling the API. It sleeps until then. -
3
Resume
At the reset it runs
claude --resumeinto the same session. If Claude had already started the task, it continues instead of starting over. You get a notification when it's done. -
4
Say hi
When a window ends and nothing is queued, the worker sends one tiny prompt, so the next five-hour window opens right then and not hours later.
Details
Runs while you're away
Queued prompts can't stop to ask for approval. They inherit the session's permission mode and your allow rules. Anything else is denied, not stuck, and hi-ai tells you how many calls were denied. Permissions
The hi costs almost nothing
It runs on Haiku with tools, skills, MCP servers, CLAUDE.md and thinking switched off: 3,434 prompt tokens, about $0.0006 at API prices. That's 7× smaller than a plain claude -p "hi", at most once per window. Measurements
Stays on your machine
hi-ai only talks to the claude command you already have. Your queue, state and logs live in ~/.hi-ai. No account, no telemetry, no runtime dependencies. MIT licensed.
Out of your way
One command to set up, one to remove. Your own hooks and settings are kept, and a backup is saved first. hi-ai doctor checks everything and prints the fix for whatever is missing.
Or queue work yourself
Line up tasks for when the limit lifts, in any project, with the model and permissions you choose.
$ hi-ai add "write tests for src/parser.ts" --cwd ~/code/app queued 4bb708; runs after the limit resets at 10/14/2026, 2:20:00 PM $ hi-ai status worker: running (pid 26105) queue: 1 pending BLOCKED (five_hour) until 10/14/2026, 2:20:00 PM, 2h 32m left five_hour 100% used, resets 10/14/2026, 2:20:00 PM (in 2h 32m) seven_day 56% used, resets 10/16/2026, 3:00:00 PM (in 61h 4m) $ hi-ai show 4bb708 ✓ 4bb708 done 10/14/2026, 2:20:41 PM ~/code/app continue in Claude Code: cd ~/code/app && claude --resume 70989282-5a10-4dc6-ac08-318b7bdec272
Questions
Does hi-ai give me more usage?
No. Your limits stay exactly what your plan gives you. hi-ai makes sure the prompt you already wrote isn't lost, and that a fresh window doesn't sit unused for hours because you weren't at your desk when the old one ended.
Which plans does it work with?
It's built for Claude Pro and Max subscriptions in Claude Code, which have the five-hour and weekly windows. It works with whatever account your claude command is signed in to.
Where does the queued prompt's answer show up?
In the same Claude Code session. You get a notification with the claude --resume command, and the conversation is there when you open it. hi-ai show <id> prints the answer too.
What if Claude was halfway through the task?
hi-ai reads the session transcript. If Claude already did work before the limit hit, the queued prompt asks it to continue from where it stopped instead of repeating the request.
Does my computer need to stay on?
It needs to be awake for things to happen on time. A sleeping laptop runs nothing. The worker catches up the moment it wakes: queued prompts run, and the hi goes out if the window has ended.
What about the weekly limit?
hi-ai waits for it like any other limit and runs your queue when it resets. It skips the hi while the weekly limit is used up, because a new five-hour window can't help then.
Does it work with claude -p scripts?
Auto-queue is for interactive sessions: a claude -p script gets the limit error back and exits before the hook runs. Queue scripted work with hi-ai add instead.
How do I remove it?
hi-ai uninstall removes the hooks and the background service. Then npm uninstall -g hi-ai. Your queue and logs stay in ~/.hi-ai until you delete the folder.
Say hi to your next window.
$ npm install -g hi-ai && hi-ai setup