Imagine gives every business owner a private team of named AI agents — each with a job, a memory, a reporting line, and a seat in your team chat. They show up tomorrow knowing what happened today.
→ / Space next · ← back · P print
Answering the same customer questions. Chasing invoices. Following up with leads three days later. Briefing whoever's covering the front desk. You don't need another inbox — you need delegation.
"What are your hours?" "Do you ship to…?" — the same 40 questions, every week.
Polite reminders, status nudges, ageing reports. Nobody enjoys it, everyone postpones it.
A warm prospect from Tuesday goes cold by Friday because you were elsewhere.
Every new hire learns it from scratch because nothing is written down twice the same way.
"What happened this week?" → 90 minutes of scrolling chat, email, sheets.
Re-orders, delivery confirms, swapping out an order — small, frequent, unloved.
It's smart — for 8 minutes. Then it forgets your business, your customers, your tone, your prices, and who reports to whom. You end up re-briefing it every single time.
Open the My Agents tab. Create one. Give it a name, a one-line job, and the team(s) it should sit in. Imagine drafts its personality (its soul.md) for your approval. Approve it, drop it into a team chat, and it's ready to be @-mentioned.
"Triage Buddy", "Invoice Nudger", "Onboarding Mentor" — name it like an employee.
You write one line. Imagine drafts identity, behaviour rules, knowledge edges, escalation triggers.
Every agent has a named boss. When in doubt, it escalates to them, not into the void.
Nothing speaks in your channels until you (the owner) approve the agent. Full audit trail.
Behind the scenes each agent is four plain text files: soul.md (identity), shorttermmemory.md, mediumtermmemory.md, longtermmemory.md. You can read or edit any of them. No black boxes.
A one-page identity: who am I, what's my job, how do I talk, what do I refuse, when do I escalate.
Today (raw transcript), this week (digest), forever (durable knowledge). Compacts automatically.
Agents live inside team chats. Anyone in the chat can @mention the agent to ask it something.
Every agent has a human manager. Out-of-scope questions get handed back to that human.
You're not bolting a chatbot onto your business. You're hiring a specialist with a job description, a memory, and a boss — which is exactly how you'd onboard a human. Same mental model. Tenth the cost. Always available.
Every conversation in the agent's team chat is appended to its short-term memory, automatically. Each night, short-term distils into medium-term. Each week, medium-term distils into long-term. Nothing is lost — but nothing is bloated either.
Raw line-by-line transcript of today's chat. The agent reads this verbatim. Cleared every night after distillation.
A daily summary appended each night. Picks up patterns, themes, recurring asks, decisions made. Cleared weekly.
The durable knowledge base — your prices, your policies, your customer types, your tone of voice. Grows slowly, never forgets.
During compaction the agent sleeps for 10 minutes — @-mentions queue up, it wakes refreshed with the new memory in place. You can read or edit any layer yourself, any time.
@frontdesk — answers hours, location, services, prices, return policy. Pulls from a single FAQ doc you maintain.
@billing — drafts a polite reminder for any invoice over N days. Escalates anything over ₹X to you.
@sales — drafts day-3 and day-7 nudges in your voice. Remembers what the prospect cared about.
@mentor — walks a new hire through SOPs, answers their dumb questions kindly, flags real gaps to you.
@weekly — on Sunday night, drops a 7-bullet summary of the week's chat, decisions, and open loops.
@supplier — drafts re-order emails, tracks who confirmed what, reminds you of delivery dates.
Each one is a 60-second setup: name it, write one sentence, approve. They start sitting in the right chat the moment you approve them.
| Generic AI chatbot | Hiring a part-time human | Imagine — My Agents | |
|---|---|---|---|
| Remembers across days | No | Yes | Yes (3-layer) |
| Has a defined role | No | Yes | soul.md per agent |
| Sits inside your team chat | No | Yes | Native — @mention |
| Approval / audit trail | No | Manual | Built-in |
| Available outside hours | Yes | No | 24×7 |
| Cost per "employee" | ~₹0 | ₹15k–₹40k/mo | A fraction of a salary |
| You own the data | Usually no | — | Yes — plain markdown files |
| Set-up time | Minutes | Weeks | 60 seconds per agent |
A newly created agent is in pending state. It can't post in any chat until you (the owner) approve it.
Every agent names a human boss. When the agent isn't sure, it tags that human instead of guessing.
Send an agent to sleep with one click — e.g. for the weekend, or while you fix its memory.
Don't like something it learned? Open the memory file, edit the line, save. The agent obeys the file.
Every reply, every escalation, every approval is recorded with timestamps. Nothing is invisible.
Memory files are markdown on disk. Export, back up, delete — at any moment.
Most owners report the first three agents — front-desk FAQ, invoice nudger, weekly review — collectively reclaim 5–8 hours a week. That's a full working day, back.
5–8 hrs/wk
Across the day-one bench of 3 agents.
~3 mins/agent
Name + one-line job description → approve.
Always-on
No sick days, no context loss, no Monday catch-up.
The intangible: your SOPs finally get written down. Every soul.md and long-term memory file is, by accident, the cleanest piece of documentation your business has ever had.
Open imagine, click My Agents, name your first teammate, write one line about what they do. Approve. Drop them into a team chat. Done.
imagine.fadafad.com · built for small-business owners who'd rather delegate than do.