Create persistent AI bots that remember, run on a schedule, use their own computer, and ask before they act.
Web · macOS · Windows · Mobile
Comparing minimums and lead times for 6 packaging suppliers
Step 6 of about 40 · claude-sonnet-5.5 on your key
"Finish the supplier comparison and draft follow-ups for the top three. Don't send anything without me."
Results expected by 7:30 AM
Every bot has one conversation that never resets. Instructions, progress, results and questions all land in the same timeline, so you always know where things stand.
Text it mid-task
It reads your message, changes course and tells you what that costs in time. No new chat, no starting over.
Heading out. Compare the 6 suppliers on the shortlist: price per 1,000, minimums, lead times. Draft follow-ups for the top three. Don't send anything without me.
On it. Comparison and drafts by morning. I'll ask before anything goes out.
Drop Boxwell, they just raised minimums. Add Fernpack instead.
Done. Boxwell is out, Fernpack is in. Adds about 20 minutes. Their price list sits behind a login, so I'll park that part for you and keep going on the rest.
To the top three suppliers from tonight's comparison. The drafts are in the thread.
Sending, buying, deleting, posting. Anything with consequences stops at a card like this until a person decides.
Nobody answers in 10 minutes? The card expires and the action does not run. Silence is a no.
Juno paused here at 2:10 AM. This page needs a person.
Fernpack Supplier Portal
Sign in to see wholesale pricing.
$ ls ~/work/suppliers corrugo.pdf northbay.pdf fernpack/ comparison.xlsx $ python compare.py --sort landed_cost 6 suppliers · 18 SKUs wrote comparison.xlsx $ _
Each bot gets a computer with a browser, a shell, files and a desktop. It runs while yours sleeps. Open it any time to see exactly what it's doing.
Take control, then hand it back
A bot's memory is a list of plain sentences, not a black box. Edit any line, see who changed it and when, and roll back a revision.
Nothing hidden
If Juno remembers it, it's on this list. If you delete it, it's gone.
Quote prices in USD per 1,000 units.
Edited by Juno · Oct 16, 3:30 AM · v3Boxwell is off the shortlist. They raised minimums.
Added by Juno · Oct 15, 9:41 PMNever send email without asking first.
Added by you · Oct 2The weekly supplier report goes to Dana on Mondays.
Added by you · Oct 9Fernpack's portal needs a human sign-in.
Added by Juno · Oct 16, 2:10 AMQuote prices in USD per 1,000 units.
Why: two suppliers quote per unit. Normalized so the comparison lines up.
Quote prices in EURUSD.
Quote prices in EUR.
Check the inbox for supplier replies since yesterday. Update the comparison and flag anything over budget.
A routine is a prompt with a schedule. It fires on the bot's own computer, on time, whether yours is open, closed or at the bottom of a bag.
Results land in the same thread
Every run posts to the bot's conversation, next to everything else it did. Approvals still apply.
Juno worked 13 hours overnight. Here's what's finished, and the one thing that waited for you.
Supplier comparison
6 suppliers, sorted by landed cost. Fernpack added after your 9:40 PM note.
Follow-up drafts
3 emails, written and ready. Not sent.
Fernpack price list
Pulled at 7:13 AM, right after you signed in.
At 11:15 PM I asked to send the 3 follow-ups. Nobody answered, so the request expired and nothing went out. Want me to send them now?
You wake up to finished work, not a wall of logs. Results up top, files attached, and the one decision that waited for you.
It only asks when it has to
Missing information, a judgment call, or an approval. Everything else, it just handles.
Bring a key from OpenAI, Anthropic, OpenRouter or any OpenAI-compatible endpoint. Every model call is metered with tokens and cost, right down to the line.
We run the bot, not the brain
JamBots never trains models and never hosts them. Your calls go to your provider, on your account.
| Time | Model | Tokens | Cost |
|---|---|---|---|
| 6:04 PM | anthropic/claude-sonnet-5.5 | 12,480 | $0.06 |
| 9:41 PM | anthropic/claude-sonnet-5.5 | 38,902 | $0.19 |
| 11:15 PM | openai/gpt-5-mini | 4,210 | $0.01 |
| 7:00 AM | anthropic/claude-sonnet-5.5 | 51,330 | $0.27 |
| All night | 148 calls | 418,230 | $2.71 |
Smaller things that make a long night go smoother.
Connect your apps
MCP integrations with OAuth sign-in. Mail, calendars, docs, your own servers. Approvals still apply.
Subagents for parallel work
Juno can split a job across helpers and merge what comes back. Up to four helpers at a time, then one answer.
Coming later
Not yetTelegram, iMessage and SMS, and voice. Same bot, same thread, more ways to reach it.
This is a draft and it will change. Every plan lets you bring your own key, so model costs go straight to your provider.
Free
Price: DraftFor trying JamBots with your own model key.
Pro
Price: DraftFor one person who wants bots working every day.
Team
Price: DraftFor teams that share bots, memory and approval rules.
When can I use it?
JamBots is in development. Join the waitlist and we'll email you when your invite is ready.
Does it work if my computer is off?
Yes. Each bot runs on its own cloud computer, so routines and long tasks don't need your laptop.
What does it do without asking?
Reading, research, drafts and busywork. Sending, buying, deleting or posting waits for an approval card.
Which models can I use?
Models from OpenAI, Anthropic, OpenRouter, or any OpenAI-compatible endpoint, on your own key.
Do you train on my data?
No. JamBots doesn't train or host models. Your bot's memory is yours to read, edit and roll back.
What will it cost?
Pricing is a draft. There will be a free plan, and every plan supports your own key.
Join the waitlist. We'll send one email when your invite is ready.