If a system has API docs, an agent can connect to it.

Tell the integration agent what the system is for and attach its documentation. That could be your field-service software, accounting platform, help desk or an internal system. It drafts the connection with every call switched off. You check where it sends data and enter the key yourself.

Illustration of connecting a new system. You describe Pellham Supply, a parts supplier, and attach its API documentation. The integration agent drafts the connection with four calls, all switched off. It then waits for you: you type the domain you expect it to use, from memory, and it confirms the connection goes where you expected before you approve it. Next, you enter the key in a sealed panel, and the agent never sees it.

Integrations New connection

You

Pellham Supply, where we order parts. It’s at pellhamsupply.com, and they gave us an API key. Their docs are attached.
pellham-api.pdf
Integration agent

Drafted Pellham Supply from the docs. It signs in with an API key, and all four calls are off.
- Search partsOff
- Check stockOff
- Order statusOff
- Place an orderOff
Waiting for you

What domain do you expect Pellham Supply to use?

pellhamsupply.comFrom memory

Goes where you expected. Uses pellhamsupply.com.

Approve Pellham Supply Show every address
Next, you enter the key in a sealed panel. The agent never sees it.

An illustration. Pellham Supply is an example.

What switching on a new connection takes

The agent reads the docs. It works out how the system signs in and drafts each call you’ll need, with the addresses it sends to.
You check the destination. Enter the domain you expect the system to use. Standard Procedure compares it with the draft’s addresses before you approve the connection.
You enter the key. It goes into a sealed panel and is stored encrypted. Agents only ever know the connection by its name.
You switch calls on. Each one asks you first until you decide an agent can make it on its own.

Built-in connections for email, records, payments and more.

Connect your account or enter a key to use these tools in a procedure. Calls require approval by default. You decide which agents can make on their own, including calls through connections built from API docs.

Email and calendars

Gmail: Search the mailbox, read threads, write drafts and send.
Outlook Mail: Find email, draft and send, file it in folders and tag it with categories.
Google Calendar: Find busy times, and create, change or cancel events.
Outlook Calendar: List events, and create or cancel them.

Documents and files

Google Drive: Search files, make folders, and copy or move what’s there.
Google Docs: Read and search documents, start one from a copy, and fill in or replace text.
OneDrive: List files and folders, and copy files.
SharePoint: Find sites, search files and lists, add list items and move files.
OneNote: Read notebooks and pages, and add notebooks and sections.
Notion: Search the workspace, query databases, and create or update pages.

Spreadsheets and records

Google Sheets: Read ranges, append rows, update cells and add worksheets.
Excel: Read ranges and tables, append table rows and build new tables.
Airtable: Find, create and update records, and comment on them.

Chat

Slack: Post messages, read a channel’s history, find someone by email and open channels.
Microsoft Teams: Send chat messages, reply in channels, find team members and create channels.

Tasks, tickets and code

Jira: Search, create, assign and move issues through their workflow, link them and comment.
GitHub: Read pull requests, open and merge them, and check workflow runs.
Microsoft To Do: Create, update and complete tasks across lists.

Payments and signatures

Stripe: Look up customers and invoices, take payments, start subscriptions and issue refunds.
DocuSeal: Send a document out for signature on your own DocuSeal server, and hear back the moment it’s signed.

The operations examples use Simpro and ServiceM8; finance uses NetSuite and Xero; support uses Zendesk and Intercom. An integration agent builds these connections from their API docs, following the same setup above.

Five ways a procedure starts.

Each procedure lists its own starts, written out as sentences, so anyone can read what sets it off.

By hand: Anyone allowed to run it starts it from its page.
By email: A message arrives that matches the sender or subject you set. That can be a shared inbox like service@, or your own inbox once it’s connected.
From an event in Standard Procedure: A client signs a proposal, or another procedure finishes a piece of work.
On a schedule: Every Monday morning, the first of the month, or whatever the job needs. Pause it or run it early from the same row.
From a signed webhook: Another system posts to the procedure’s webhook address and signs each message with a secret you both hold. Unsigned messages don’t start anything.

Illustration of a repair procedure’s How it starts list, with five ways in: started by hand, an email to the service inbox, a client signing a service agreement, every Monday at 7:30, and a signed webhook from the parts supplier.

Repair request How it starts

When someone starts it from the procedure’s pageAnyone allowed to start runs on request
When an email to service@ mentions “repair” in the subjectShared inbox, or your own once it’s connected email
When a client signs a service agreementSomething that happens in Standard Procedure trigger
Every Monday at 7:30Next: Monday 12 October schedule
When the parts supplier sends a signed webhookAnything else that can post a signed message webhook

An illustration. The repair business is an example.