A database for your team and your agents.

You build types such as item, purchase order and invoice, link them to each other, and agents read and write in them, with a revision for every change.

What are records?

Records are tables you define yourself: one type per thing, one row per entry, fixed field types per column. Unlike Excel, rows are linked to each other and to documents, and every row knows its revision.

  • Skills read and write

    A skill queries records with filters and writes results back, for example the suggested order as a new purchase order.

  • Workflows react and fill

    A Shopify webhook creates one row per sale. A schedule checks stock against minimum stock every night.

  • Documents hang on the row

    The invoice PDF is linked to the record. From the amount in the table you are one click from the document.

  • Open to the outside

    Your own agent reads and writes over MCP, and every view can be exported as CSV.

From the type to the checked delivery

  1. 01

    Build types with fields and rules

    Fixed field types, required fields, unique values. An agent may prepare the draft, only you can publish.

    • „Create a type "Purchase order": unique number, supplier, status, total, line items.“
    • „Add a required field "Minimum stock" to "Item".“
  2. 02

    Relations between the types

    Relations are real links in the database. An invoice cannot point to a purchase order that does not exist.

    • „Link every invoice to its purchase order by the order number.“
    • „Show on the purchase order how many goods receipts are attached.“
  3. 03

    A purchase order over its whole lifetime

    Quote, invoice and goods receipt hang on the same record. That is why it shows when 150 units were invoiced but only 140 delivered.

  4. 04

    Writes that hold

    Every type is its own PostgreSQL table, every write a transaction. A batch runs completely or not at all, and concurrent changes never overwrite each other.

  5. 05

    Every change with a revision

    Nothing is silently overwritten or deleted. Every earlier revision can be restored, a whole agent run reverted.

Permissions

What an agent may do and what only you may do

The line is set in the platform, not in the prompt. For these actions an agent gets a refusal and has to ask you.

  • Agent: read and write

    Query, create, change, link and archive records, within the types and permissions of the space.

  • Agent: drafts

    Prepare types, fields and relations as drafts. The preview shows what publishing would change.

  • Only you: publish and delete

    Publishing types, permanent deletion, large imports and restoring snapshots are reserved for people.

  • Limit per run

    One run of a skill or workflow may touch at most 5,000 records. A mistake in a skill cannot rewrite the whole table.

Common questions

No. Shopify stays the shop, the warehouse system stays the warehouse system. Records hold what lies in between and otherwise ends up in Excel: purchase orders to suppliers, the match with invoice and goods receipt, open credit notes.

Every change names the revision it builds on. If the record has changed in the meantime, the second change is rejected and receives the current state. An agent then reads again and retries. Nothing is overwritten.

In through workflows (for example a Shopify webhook per order), through agents, through a CSV import or by hand in the table. Out through a CSV export with the same filters as the view, through workflows or over MCP.

Yes. You change the draft and publish again. If a field type is converted, the platform takes a snapshot first. Values the new type cannot hold can be brought back from it. Deleted fields sit in the trash and can be restored.

Describe the task. The rest takes shape on the platform.

Start for free, create your first skill with the assistant, and see what a run looks like before you connect anything.

Start for free