Wrasse lives in the app you already have open.
A personal AI agent on Telegram, Slack or the Wrasse app. We host it, it arrives tuned for the work you do, and every tool call it makes is checked before it runs and logged where you can read it.
Same agent, whichever app you live in.
The four things off your desk.
Follow-ups, drafted
Leads that have gone quiet get flagged with a nudge already written. Approve a batch with one reply.
Showings, booked
Holds the slot, confirms with the client, and puts it on your calendar before the request goes cold.
Comps, pulled
Formats the numbers into a one-pager you can forward without touching it.
The pipeline, watched
Tells you what moved, what is stuck, and who to call first tomorrow.
From morning rush to evening wrap.
You don't manage the agent. You delegate, and it reports back — with every tool call on the record.
- 7:40 AM
Three leads went quiet overnight
Flagged with a nudge already written for each one. You approved all three over morning coffee, with one reply.
- 11:15 AM
A buyer asked for comps on Henderson
Pulled them, formatted a one-pager, and held it for your read before anything went out.
- 2:00 PM
A showing request came in
Held the slot on your calendar, confirmed with the client, and finalized it before the request went cold.
- 6:30 PM
End-of-day recap
What shifted, who is waiting, and who to call first tomorrow.
Enterprise agents are built for org charts. This one is built for you.
One person, one agent, one thread. No seats to provision, no admin console, no rollout.
Open source
Total control, if you have Docker, a spare weekend, and the appetite to audit a skill catalog yourself.
Wrasse
We host it, we curate the skills, and it arrives already knowing your trade. It acts inside limits you can see, on a machine of its own.
Enterprise platforms
Powerful, and priced per seat with a procurement cycle attached. Excellent — for a company that has one.
Every tool call, on the record.
Reading your own message threads is always allowed under your default policy. No approval needed, and it is still written to the log.
Read-only lookup against the MLS connection you authorised at setup. Nothing was written or shared.
A rule you set holds messages to this client for approval. The draft sat in the thread until you replied “send”.
Released by you at 07:44 with a one-word reply. The exact text that was sent is stored against this entry.
Calendar holds on your own diary are pre-approved. Invitations to other people are not — that is a separate rule.
Open any row to see why the agent did it.
Private by architecture, not by promise.
An agent that reads your messages and acts for you has to earn trust in its design, not its marketing. This is how Wrasse is built:
-
Its own machine.
Your agent runs on its own virtual machine with its own volume. It never shares a virtual machine with another user's agent.
-
Checked before it acts.
Every tool call passes a policy check before it runs, anything your policy holds asks you first, and every tool call lands in your audit log.
-
Model keys never reach the agent.
Model-provider keys live in the control plane, outside your container. The agent asks the control plane, and never sees them.
-
No training on your data.
Not your conversations, not your documents, not your activity.
-
Stored credentials, encrypted.
Third-party tokens, integration keys and platform secrets are stored encrypted with AES-256-GCM.
-
Closing your account schedules it for erasure.
Your agent stops at once, and your data is erased 30 days later, apart from what the Privacy Policy lists as kept or not yet erased. Backups age out on their own schedule.
The full breakdown, including exactly where we stand on certifications →
Cleaner wrasses are reef fish that run cleaning stations. Larger fish voluntarily approach, queue up, and hold still while the wrasse picks parasites from their gills. The relationship is so reliable that fish remember individual stations and return to them. That's the model — a small trusted agent that bigger systems come back to because the work it does makes them work better.