An assistant inside your scope
It searches, summarises, drafts replies, and proposes a task or a note — you approve the proposal. What a colleague on "own records only" cannot see, their assistant cannot either.
AI and integrations
The assistant reads with the same access your screens do, and writes nothing by itself: it proposes, you press the button. Outwards: a REST API, signed webhooks, an MCP server and connected apps.
It searches, summarises, drafts replies, and proposes a task or a note — you approve the proposal. What a colleague on "own records only" cannot see, their assistant cannot either.
A step writes text or sorts into the company's own categories; "Write with AI" sits in the website, blog, email template and product description fields, and nothing goes out until you save.
Prepaid credits, charged at the cost the provider actually reports, with optional autopay and a monthly ceiling — or your own Anthropic key, in which case no credits are taken.
Scoped keys, an OpenAPI document, signed webhooks with retries and a delivery log, and an MCP endpoint where an AI agent works by the same rules as the screens.
Slack, Microsoft, Google, HubSpot, Notion, Airtable, Stripe and more — by OAuth or a key, with an automation step that acts in them. A credential can only be sent to its own service's address, and is stored encrypted.
A note or an inbound email was written by somebody else, so it is fenced and neutralised before the model sees it. A question runs at most six rounds of tools, and every request counts against the plan's monthly AI allowance.
The same thing step by step, in the help centre.
With your own key, Anthropic's models; on the platform's key the call goes through OpenRouter, which is where credits are accounted. With AI switched off, every AI feature stops.
The core CRM is free for good, with no card; you can try every paid module for 14 days. If it doesn't suit you, export everything and leave — the data stays yours.