AI agents that answer from your own knowledge, and then do the work
Prospects, clients and your own team reach you through one platform, your portal, website, email and phone, where AI agents answer from your company's own knowledge and then do the follow-up work. A better experience for everyone you serve, with the team you already have.
What it does
Six kinds of work AI agents take off your team
Pick the one that sounds like your week.
Document & data processing
Attachments are read the moment they arrive, the figures that matter go into the shape your ledger expects, and anything uncertain goes to a person with the source page beside it. At month end the report writes itself from your template.
- Attachments captured straight off inbound email
- Uncertain values sent to a reviewer
- Reports from your own templates
Inside the platform
The five parts you work with every day
Five pieces carry every agent and automation you build.
Draw the work once, on one canvas
Chat agents, voice agents and unattended automations are built on the same canvas, from the same pieces: an agent with its instructions and knowledge, a tool that acts, a decision, a loop, a review step. Every change is versioned, and a live automation never falls back to a half-finished draft.
- One canvas for chat, voice and automations
- A review step wherever you put one
- Versioned, and published all at once
MCP, the standard way agents connect to other systems, lets you choose method by method exactly what an agent may do
The accounting system, the CRM, the document storage and the calendar you already use, connected with credentials you authorise yourself — encrypted under your own key and decrypted only at the moment of use. Through MCP, you choose method by method exactly what an agent may do in each of them.
- Your own credentials, under your own key
- Permissions chosen method by method
- In-house systems through a plain request
Every client, in one record
Contacts, the company each one belongs to, the meetings you held with them and the tasks those meetings produced — on one timeline, beside the agents that serve them. Each account has a person in charge on your side, administrators on theirs, and a usage allowance you can cap.
- One timeline per client
- A person in charge for every account
- Usage allowances you can cap
Meetings that write themselves up
Record an in-person session from the dashboard or the mobile app. It is transcribed live with speaker labels, summarised from your own template, and the tasks it produced land on the client’s timeline before the next meeting starts.
- A live transcript with speaker labels
- Summaries from your own template
- Tasks on the client’s timeline
The parts that make it safe to run
Roles and permissions you configure, sign-in through the work account you already have with a second factor where you require one, and standing instructions that apply to every agent at once — previewed before they apply, and rolled back in one move. The interface ships in five languages, and agents reply in the language they are written to.
- Roles and a permission matrix
- Standing instructions for every agent
- Five interface languages
Plenty of tools answer. This one does the work.
The first two are hard to match anywhere else: reaching systems with no API, and being safe to leave running. The rest are why a business can run it without an engineer.
- Reach
It reaches the systems that have no API
Most tools stop where their integrations stop. A hosted browser signs in to a supplier, carrier or government portal with your own credentials, confined to destinations you authorised, and every action is recorded as evidence.
- Only where you authorised it
- Recorded as evidence
- Safety
It is safe to leave running
An action that may or may not have completed is held for a person to reconcile, never silently repeated, and sending limits stop rather than fail open.
- Held, never repeated
- Limits stop, not fail open
- Action
It acts, not just answers
It books the meeting in a real calendar, updates the CRM, drafts from your template, files where your team already looks, posts to the ledger and routes the escalation, and it asks a person first whenever an action has consequences.
- Books, drafts, files and posts
- Asks before anything with consequences
- Trust
Grounded and governed
Answers come from your own content with the source cited, your data is isolated from every other customer's, and every interaction, tool call and escalation is on the record and exportable.
- Cited from your own material
- Every action logged and exportable
- Control
Run by the people who run the practice
A manager configures it without engineering, on one canvas. Every change is versioned, publishing is atomic and a change that turns out wrong rolls back.
- No-code, on one canvas
- Versioned, with rollback
- Surfaces
One configuration, every surface
Website widget, branded portal, dashboard and mobile app, email and phone line, with the same agent, knowledge and rules behind all of them.
- Portal, widget, email and phone
- One knowledge base behind all of it
The gap we close
Most AI tools stop at the answer
Between "AI that talks" and "AI that does the work" there is an unserved middle: agents that understand a company's own knowledge and are connected to its real systems, with the guardrails and audit trail a business needs to trust them.
Chatbots talk
Generic assistants answer, but they cannot act. They do not touch the calendar, the ledger or the CRM.
Automation acts
Traditional tools act, but they do not understand — rigid rules with no grasp of a company’s knowledge or the nuance of a conversation.
Agents close the loop
The agent takes the next action a person would otherwise have done by hand — grounded in your own content, inside your own systems.
Before you decide
What happens to your data
Specific claims, not a badge, because the specific ones are the ones we can stand behind.
Your data stays in the EU
Not a policy — a constraint built into our infrastructure. A deployment outside the EU fails our own build checks, verified automatically on every change.
Your own encryption key
Each customer gets a dedicated key, created when their account is set up. Credentials and meeting transcripts are encrypted under it.
Strict separation between customers
Each customer’s records and search index are isolated; one customer’s information cannot surface in another’s answers.
Your data trains no one’s model
AI providers are configured so customer content is never used as training data.
Everything an agent did is on the record
Every interaction, tool call and escalation is logged and exportable.
Every upload is scanned
Files are scanned for malware before they are accepted. If the scanner is unavailable the upload is refused rather than trusted.
Who it’s for
Companies that hold real knowledge and repeat real work
Small and mid-market companies running high-volume, repetitive knowledge-and-action work.
Health & wellness practices
Bookings while you're with a client, and consultation write-ups that don't wait for the evening.
Read moreSee it answer, and then act, on your own work
A short conversation is the fastest way to tell whether this fits your business.