Connect
Fathom for your calls, Cal.com for your calendar, HubSpot for your pipeline, Brevo to send. Every permission is read only, except the one that sends a message you approved yourself.
Five agents read your calls, your calendar, your pipeline and your own playbook. Everything they write lands in one queue. Nothing goes out without you.
Three to twenty people. A CRM nobody spent six months configuring. Calls recorded, notes taken, and follow-ups that slip because the week moved on. That gap is where the agents work.
Fathom for your calls, Cal.com for your calendar, HubSpot for your pipeline, Brevo to send. Every permission is read only, except the one that sends a message you approved yourself.
The agent pulls your positioning and your real emails from the context before it writes a line. Post-Call Follow-up runs at noon and 5pm on weekdays, CRM Watchdog at 8:30.
The draft arrives with its trace: what was read, which tools answered, what the run cost. You edit, you send, or you bin it.
One real run, from click to send: the agent runs, the draft waits for you, you edit, you send.
Agents
Each agent reads your calls and your context, then writes a draft. None of them ever sends anything.
Reads the notes from recent calls and writes the follow-up, in the founder's voice.
You do not have to take our word for it. Open any draft and you get the log of the run that wrote it: which meetings were opened, what was looked up in your context, what came back, and what it cost.
| Action | What came back |
|---|---|
| Read recent callsFathom | 3 calls, 1 with no follow-up |
| Open the meetingFathom | northbase · 42 min · 6 action items |
| Search your GTM contextNotion | founder voice, Lumen story |
| Write a draftDrafts | 1 draft, waiting for you |
In plain words: the agent read 3 recent calls, found one with no follow-up, checked your voice and your customer stories, wrote one draft, and stopped there.
Example for Kalio: only the steps that use the connectors on this account. Not an exhaustive list of what agents can do.
One for each moment of the cycle, from the signal you never saw to the deal going quiet. Every one of them writes a draft, and not one holds the send button.
Looks for the buying signals of your ICP: a role being hired for, a raise, a change at the top, the trigger event you sell against. Writes up to five briefs: company, signal, angle. Never an email ready to send, because it knows a company, not an address.
Reads the next days of bookings, digs into the company, the persona and the likely pain, then leaves you five questions in the order to ask them.
Reads the recap, writes the follow-up in your voice, and picks the customer story that genuinely matches the profile.
Takes the recap and your own offer template, drops in the prospect's numbers and the scenario that fits them. Still to be built in the product.
Sorts your open deals by days of silence, from the friendly nudge at three days to the walk-away message at twenty-one.
12 tools are live. The rest of the catalogue is planned, and we would rather say so than show you a dead button.
Planned
Triscale GTM AI Agents reads the tools you connect and writes drafts that a human approves. Here is what we ask a Google account for, and why.
calendar.events.readonlygmail.sendgmail.readonlydrive.readonlyopenid, emailData hosted in Paris, never sold, never used to train a model. Privacy policy
Six limits that live in the code and not in a sales promise. The first one is the reason the other five exist: no agent tool can reach the sending function.
The rule lives in the code, not in a promise: no agent tool can reach the sending function. A human can ask the platform to send a specific draft they just read. An agent cannot.
Cal.com is booking read. HubSpot is contacts, companies and deals, read only. Fathom is its single public API scope. Gmail is send plus a search of threads with a known prospect. Brevo's write scope is used to send what you approved.
Connection tokens are encrypted with AES-256-GCM under a key held on the server, never returned to the browser, never written to a log. Passwords are hashed with bcrypt.
Every database query is scoped to your account. Deleting an account deletes its connections, drafts, runs, notes and context with it.
Every run keeps its trace: tools called, inputs, outputs, errors, token cost. When a run produces nothing, you can see exactly why.
Your data is not sold, not used for advertising, and not used to train models. We look at your content only when you ask us to look at a problem.