Connected systems

Four systems have working adapters. Twenty-two do not, and here they are.

The connection vocabulary describes twenty-six business systems across six categories. Four of them have an adapter that can actually answer a question today. This page is the whole catalogue rather than the working part of it, because the number that decides whether this is useful to you is not four, it is whether your system is one of the four.

Planning reads 1/1 because the category has one system in it, not because planning is finished. The bars are how complete each category is against its own contents, so a full bar on a category of one is a smaller claim than a fifth of a bar on a category of five. Finance and Service are drawn because they are declared, and they are empty because nothing in them has been built.

The catalogue

All twenty-six, in the order the code declares them.

A tool asks for a category rather than a product. It calls crm_read, never hubspot_read, which is what lets a second adapter in the same category answer the same tools with nothing downstream changing. That is why this list is worth publishing before most of it exists: the shape of the work is already fixed, and the tools that would serve each category already have names.

System Adapter How a read signs in
ActerysPlanning, over SQLliveone shared credential
HubSpotCRMliveone shared credential
SalesforceCRMmodelledeach person's own token
Microsoft Dynamics 365 SalesCRMmodelledas the signed-in person
PipedriveCRMmodelledeach person's own token
Zoho CRMCRMmodelledeach person's own token
ClickUpTasksliveeach person's own token
JiraTasksmodelledeach person's own token
Azure DevOpsTasksmodelledas the signed-in person
AsanaTasksmodelledeach person's own token
monday.comTasksmodelledeach person's own token
LinearTasksmodelledeach person's own token
SharePointDocumentsliveas the signed-in person
ConfluenceDocumentsmodelledeach person's own token
Google DriveDocumentsmodelledeach person's own token
BoxDocumentsmodelledeach person's own token
DropboxDocumentsmodelledeach person's own token
NotionDocumentsmodelledeach person's own token
XeroFinancemodelledone shared credential
QuickBooks OnlineFinancemodelledone shared credential
Microsoft Dynamics 365 Business CentralFinancemodelledas the signed-in person
NetSuiteFinancemodelledone shared credential
SAP S/4HANAFinancemodelledone shared credential
Sage IntacctFinancemodelledone shared credential
ZendeskServicemodelledeach person's own token
ServiceNowServicemodelledone shared credential

For the four that are live, the sign-in column is enforced. For the other twenty-two it is an expectation. The platform deliberately does not refuse a different arrangement on the strength of a guess: a client whose Salesforce is genuinely operated through one integration account would otherwise be told their own setup is impossible by a product that has not written a line of Salesforce code. A kind with no adapter is not pre-judged, and a test holds that open.

The column that matters

Three ways a read signs in, and only one of them stores nothing.

That third column is not an implementation note. It decides whether your own system applies your own permissions, or whether our grants are the only thing standing between a person and the data. Two questions produce every row of it, and neither answer is ours to choose: can the far system exchange our caller's identity for its own, and does its API then enforce that person's permissions?

Eight of twenty-six have nothing behind us. That is the row a security review should spend its time on, and it is the same argument as whose permissions apply, counted across the catalogue rather than across the four systems that work today.

As the signed-in person

The person's own sign-in is exchanged for access to the far system, and nothing is stored here at all. The far system then enforces its own permissions per item, so a document they cannot open is a document they do not get, regardless of anything we decide. This is the mode worth wanting, and four systems support it.

Each person's own token

The system enforces per user but offers no way to exchange an identity, so each person supplies their own credential. Real per-person visibility, bought with a custody problem fourteen times larger than one secret: it is one stored credential per person, revocable on its own, usable by nobody else. Better than shared, worse than storing nothing.

One shared credential

The system offers nothing better, so every caller reads through the same identity and the system cannot tell them apart. Our grants, the tool list and your own row rules are then the entire access control. We say so on the screen where somebody decides whether that is acceptable, rather than in a footnote afterwards.

One change unlocks the largest block of this catalogue, and it is not an adapter.

Every system that can read as the signed-in person sits behind Microsoft Entra: SharePoint, Azure DevOps, Dynamics 365 Sales, Business Central. They all come off one on-behalf-of exchange. So the sequencing fact worth more than any individual connector is that a single change to the sign-in broker makes four systems possible at once, in the only mode that stores nothing.

What modelled means

A declared system is a shape, not a date.

Being in this catalogue means the category, the expected sign-in and whose rules would apply are already decided, and that a connection row naming it can be saved. It does not mean the read works. Opening a connection whose adapter does not exist is refused readably rather than handed to a reader that would misread it.

What a listing gives you

  • The category is fixed, so the tool names that would serve it already exist.
  • The sign-in mode is worked out, which is the part that takes the argument.
  • Whose narrowing applies is stated before anybody commits to the arrangement.
  • The console tells your admin which is which, rather than accepting a connection that would fail the first time somebody used it.

What it does not give you

  • No date. None of the twenty-two has one, here or anywhere else.
  • No working read. A modelled system answers nothing today.
  • No promise about the mode. For an unbuilt system it is what we expect, and building it may prove us wrong.
  • No queue position. What gets built next is what a client is paying for, which is usually not the most popular system.

One boundary in this list is permanent rather than unfinished. Every Finance adapter is read-only for good, and not by a switch somebody can flip. A mistaken write to a ledger is a different class of harm from a mistaken write anywhere else, so a legitimate future need to draft an invoice has to become a code change. That is the correct amount of friction. The rest of the limits are on what it cannot do.

Is your system one of the four, or one of the twenty-two?

Worth five minutes on a call rather than a guess from this page. If it is one of the twenty-two we will say what building it would actually involve, including when the answer is that you should not wait for us.