An assistant that does the work — and shows its work
Ask for what you need in plain language. The Assistant proposes an action, shows you exactly what will happen, waits for your approval when it matters, and records what it did. Sales, support and operations in one workspace.
Prices are published at launch; every account starts with a trial. No card to start.
What it is, and what it is not
It is a working assistant for a company: it reads your CRM records, tickets, projects, invoices and documents, drafts customer emails in your voice, proposes the next step, and — with your approval — carries it out through your connected systems. Every action is planned first, previewed, and stored with who approved it and when.
It is not an autopilot. Nothing external — no email, no calendar invite, no payment, no deletion — is sent because a model suggested it. Those actions stop at a preview and wait for a person. Financial and destructive actions have their own approval levels, and role permissions are enforced on the server, not in a menu.
Built for three kinds of work
Sales
Customers, deals and a pipeline by stage. Ask who has gone quiet, get a follow-up drafted from your own past messages, approve it, and it goes out from your own address. Read more
Support
Tickets with states, priorities and response deadlines; an inbox review that notices who is waiting on you and proposes replies you approve. Read more
Operations
Projects, milestones, tasks, schedules and multi-step workflows that run on a schedule and stop for approval where the rules say so. Read more
How a request becomes a result
You ask
In your words. The Assistant reads only the company facts relevant to the message, within a fixed limit — never your whole database.
It proposes
One answer, and at most one action from the list your role allows. The list is decided by the server; the model cannot invent an action.
You see the preview
Exactly what will happen, built from the stored plan — the same record that will execute — with its risk class and approval level.
It runs, or waits
Read-only lookups run at once. Anything that writes, sends, pays or deletes waits for the approval the rules require, then runs through the verified connector.
It is recorded
The action, the approval, the result and the model cost are stored per workspace, visible in Actions and in Usage and Costs.
Integrations, stated truthfully
Connections exist for Gmail, Google Calendar and Google Contacts, HubSpot, Slack, any SMTP/IMAP mailbox, outbound webhooks and Stripe Connect. Each one shows its real state in your workspace — connected, or not yet — and nothing is marketed as working until it is connected for you. See the integrations
Security and cost, visible
Workspaces are isolated by organisation on every query. Approvals are bound to the exact plan they approve. Model usage is measured per call — model, route, tokens and cost — and enforced against your plan's limits on the server. You can see it in Usage and Costs at any time. How security works
Questions people ask
Will it send emails on its own?
No. It drafts; a person approves; then it sends from your own connected address. There is no code path that sends without an approval record.
Which model does it use?
A cost-appropriate one per task: a small model for classification and extraction, a stronger one for conversation and drafts, and the most capable model only when a task explicitly escalates and your plan includes it. Every call is recorded.
Can I try it without a card?
Yes. Every workspace starts with a trial and no payment details. Prices are published at launch; until then the plan pages describe what each plan includes.
Where is my data?
In your organisation's workspace, separated from every other workspace on every query, with role-based access enforced by the server.