Connect · Data connections
Connect what you run. Nothing to copy first.
Databases, systems with a web API, ready-made connectors and your Microsoft 365 files, read where they live, each person as themselves where the system allows it. Point it at a new API and the assistant works out how it is used, then asks you to approve what it keeps.
- Read the API's own description41 paths
- Made 12 read-only test callsnothing written 1Read-only test calls
- Checked how each list pages
GET /v2/work-orders1,284 recordsPaging checked
2Paging proven, not guessedGET /v2/techniciansPaging checkedGET /v2/sitesPaging checkedGET /v2/invoicesNeeds another sign-in- 1Read-only test calls
- 2Paging proven, not guessed
- 3You approve what it keeps
Three ways in.
1 · Databases
PostgreSQL, MySQL, SQL Server and MongoDB, read where they are: tables, typed columns, keys and the links between tables.
2 · A system with a web API
Its address, how to sign in, its lists, where the rows sit and how its pages turn. "Test connection" makes a real sign-in and shows the system's own reason if it fails. Eight sign-in settings for a web system, from none to API keys, tokens, username and password, and OAuth 2.
3 · Ready-made connectors
For Xero, QuickBooks, Odoo, HubSpot, Stripe, SendGrid, Notion, Airtable, GitHub and OpenWeatherMap. Pick one, type your key, and its lists arrive set up; the key is tested before it is saved. Stripe, GitHub, HubSpot and SendGrid carry the vendor's whole published API. Xero and QuickBooks first ask your company to register an app with them.
And your Microsoft 365 files: your company's owner connects its SharePoint sites, and each person can add their own OneDrive. Plus the documents people attach in the chat.
If your system keeps its data in PostgreSQL, MySQL, SQL Server or MongoDB, has a web API you can sign in to, or is on our list of ready-made connectors, it can be connected. In the demo we connect yours and you see how it answers.
Xero and QuickBooks first ask you to register an app with them.
It learns a new API with you.
Point the assistant at a system's API, and it works out how the API is used before anything is added.
read-only test calls
- 1Reads the API's own description. It looks for the published description (OpenAPI or Swagger) at the address you give, or in the usual places.
- 2Makes safe test calls. Read-only, never a write. Private and internal addresses are refused.
- 3Tries values to learn what each list is for. It varies the parameters, opens the detail of several records, and checks the fields against the system's own description and the values that come back. When the system refuses a parameter, it reads again without it and says so.
- 4Proves how each list pages. It asks for a large page to find the most the server gives, fetches a second page to prove the pager moves, and says how many requests a full read takes.
- 5Checks an empty answer. A list that returns nothing is compared with queries your company already runs.
- 6Proposes; you approve. Up to 20 lists that answered, as tick boxes. Each is added on a card you can correct, and a preview then proves it returns rows.
- 7Keeps what it learned, in your words. What worked is written to the connection's "What the assistant knows" page, where your people keep what each list means in their own words, and can correct it.
I read its description and made read-only test calls. 4 lists answered.
data.items
Pages bypage · page_size
Needsstatus
| Job | Site | Status |
|---|---|---|
| J-20817 | Quellmark Foods | open |
| J-20816 | Lumora Hotels | HLD |
| J-20811 | Fenwright Builders | open |
- 1Nothing added until you approve
- Already connected, list missing? "Add data" finds the list across your connections, asks for a missing value, and when the list exists on the API but is not added yet, investigates it and proposes it.
- Have an OpenAPI address? Import its lists, with a preview first.
- Connect a ready-made connector or a public API by asking. You type the key on a card, never in the chat.
- In July we ran it end to end on a real, thinly documented, self-hosted engineering tool, with fixes made along the way. It found the endpoints and how each is used, chose the lists to keep history of, set up the copies and built dashboards on them.
- It does not run on its own: it works when a person asks, and nothing is added until a person approves.
What your data means, in your people's words.
- One page per connection, "What the assistant knows". For each list: what one row is, the words your people use, what they normally mean by it, the field people call a row by, what to show, the traps, and what each coded value means. Whoever may edit the connection can read and correct it. It grants no access.
- Lists get roles: core, supporting, internal, admin or retired, and the departments they serve. The assistant reads core lists first, admin lists only for administrators, and never a retired one; it names the one to use instead.
- Usage suggests, a person confirms: "This list is read 41 times by Sales. Make it core for Sales?", with what will happen, who it affects and how to undo it.
- Teach it in the chat: "remember this about the connection" becomes a card. Once approved, it joins the connection's knowledge.
- Live or test copy: each connection is marked, so the assistant stops asking "which one?" and offers risky actions on the test copy.
- Some connectors arrive with their meaning prepared. Odoo's ready lists come with codes in words, currencies and time zone. An older connection is brought up to date in one click.
- A new system starts with its list and field names; we add its meaning with you.
One row is one visit to a site. People say 'job' or 'call-out'.
Call a row by its job number.
Status 'HLD' means on hold, waiting for parts.
1In your people's words- 1In your people's words
Each person as themselves.
each person as themselves
- Where a system allows it, each person connects with their own account, so the assistant sees only what that person may see in that system. A database, or a system with one shared account, is read as that one account.
- This works on any web system, not only with OAuth: a person's own token, own login or own Odoo login. A person who has not signed in is refused, never served the shared account.
- My access shows every account a person has linked, in words: "Connected as …, since …", or "Needs you to sign in again", and why. Connect, reconnect or disconnect from there.
- Sign-ins are typed into a form or a card, never into the chat. They are kept in your company's own vault and never shown back, owners included.
- On a shared system, an owner can let each person see only their own rows, their team's, or their branch's. This works on web systems, not on a database connection.
- Share a connection with people, teams or roles, to view, edit or delete. Hand a leaver's connections to a colleague in one go.
Xero and QuickBooks first ask you to register an app with them.
- 1Never shown back
Gentle with your systems.
It reads your systems when you ask. There is no data warehouse to build first, and nothing to move before the first question.
A question reads only the lists it needs, when it is asked, and at most 20,000 rows of each. To a web system it sends a few reads at a time and queues the rest, so a busy system is not sent a burst.
On a web system, a list it has read is reused for about two minutes, then read again. That memory is kept apart for each person's login, and any write clears it.
A system's sign-in is kept between questions, instead of signing in for every one.
If you would rather questions did not reach your live database, connect a copy of it, with an account that may only read. Answers are then as fresh as the copy.
Private and internal addresses are refused and every redirect is checked. Flows reach only the web addresses your owner allows.
Writing to a system is off for every connection until your owner switches it on and approves each kind of change.
It keeps your chats (unless you choose a private chat), the files you attach, and the dashboards and queries you save, and a history copy only of the lists you choose to track. How long each is kept is on our privacy page.
Each company's data is kept in its own databases, apart from every other company's, with a separate vault for sign-ins.
Your company profile, in five answers.
- Five short questions. An owner can answer them: what you do, how you earn, which currency you report in, which system is for what, and which words trip people up.
- The assistant uses the confirmed answers from the next question on, and only for people allowed to see where they came from.
- Add files and links as sources too. What is drafted from them stays a draft until a person confirms it, and a line goes when its source goes.
One set of connections for every engine.
The assistant's answers, dashboards, the history engine's copies, analyses, workflows and agents all read through the same connections, as the same person, under the same sharing.
Dashboards
A board can join a list from one system to a list from another by a column they share, such as a customer number. Give a connection its system's web address, and a click on a chart opens the record there.
DashboardsChanges in your systems
Only through the actions your owner declared on a connection, approved by the people named there, with a record of every call. Letting an action run without a person needs a permission only our staff hold.
Safe removal
Removing a connection first shows everything that would break: queries, boards, flows, analyses and history copies. An owner can also switch the Data app off for the company: its pages say so, and nothing is deleted.
Not in our catalogue?
We add your system as data: nothing waits for a new release.
Who uses it.
Which systems can it read, and as whom?
Each connection shows what it reads; where a system has its own logins, each person signs in with theirs.
Built with Data connectionsFor IT & data CEO or ownerDoes it work with the system we use?
If it keeps its data in one of the four databases, has a web API you can sign in to, or is a ready-made connector, it can be connected; in the demo you see how it answers.
Built with Data connectionsAI AssistantFor CEOs & owners OperationsOur field service tool has an API but little documentation. Can it read it?
The assistant investigates the API with read-only calls and proposes the lists that answered, for you to approve.
Built with Data connectionsHistory engineFor operationsWhat it does not connect yet.
So you know before the demo.
- Google Workspace: Google Drive, Gmail and Google sign-in are not built.Files and mail are Microsoft 365 only.
- Asking about your own Outlook mail and its attachments is not offered yet.
- Finding a file across SharePoint and OneDrive by what is inside it is still being proven.
- A workflow cannot read a connected database today.
- A new system does not arrive knowing what its lists mean: it starts with list and field names, and its meaning is added with you.
- The API investigation needs a person to ask and to approve, and proposes up to 20 lists at a time.It has not yet passed a scored set-up run on our live service.
- "How do I…" answers from your software's own manual, or from your company's own, are built but not yet reliable enough to offer.
Bring the system nobody has connected yet.
In the demo we connect one of your systems, and you ask the questions.