Hub / Connectors

Connectors

Five platforms, four of them shipping and one not built yet. Start here to find your platform, see how the connectors actually differ, and work out what to do if yours is not on the list.

GA Salesforce · Dynamics · Zendesk Early access Fin / Intercom Not built ServiceNow

Routing

Find Your Platform

Salesforce
A managed package installs into your org. Architecture & setup →
Microsoft Dynamics
A managed Power Platform solution installs into your environment. Architecture & setup →
Zendesk
A ZAF app plus a ZIS integration, both in your own tenant. Architecture & setup →
Fin / Intercom
TSANet hosts and operates the adapter; you install almost nothing. Architecture & platform notes →
ServiceNow
No connector yet — development committed for 2026. Platform assessment →
Anything else
Build against the API directly. See below →

Comparison

How the Connectors Actually Differ

The differences that change how you design, operate and support an integration — not feature checkboxes.

Why the classes differ →
ServiceNow columns are empty because nothing is built. The assessment covers what the platform forces either way, and lists the questions that are still open.
  Salesforce Dynamics Zendesk Fin / Intercom ServiceNow
ClassIn-tenantIn-tenantPartialGatewayUndecided
StatusGAGAGAEarly accessNot built
Operated byYouYouYouTSANet
Agent surfaceLWC on the Case record pageHTML + JS web resources on the Case formZAF sidebar app + nav-bar screenCanvas Kit panel in the Help Desk
Where state livesCustom objectsDataverse tablesTicket fields and tagsThe gateway's own durable store
Inbound deliveryApex REST on a public SiteScheduled poll by default; optional webhook to Power AutomateZIS generic inbound webhookTSANet push to the adapter
ReconciliationScheduled job, 15 min defaultScheduled flow, 15 min defaultZAF background poller, 60 sReconciliation poller
Extension pointClonable Flow libraryCase Mapping configuration tableZendesk triggers and ZIS flowsServer-side, by TSANet
Auto-acceptAsset-match flow, rejects on no matchAuto-routing by case number, then customer nameBuilt-in setting, or a scoped triggerDeliberately manual
DistributionManaged packageManaged solution / AppSourceApp upload + ZIS bundleA connected Intercom app plus a hosted service
Shipping a fixMembers upgradeMembers upgradeMembers upgradeCentral, nothing to do
Connector faultsIsolatedIsolatedIsolatedShared — one hosted service
Platform outageAffects every member in every class — the Connect API is the shared floor

Two rows are worth pausing on. Reconciliation tells you what happens when a push is missed — and the Zendesk poller runs in a browser page, so it only catches up while an agent has a tab open. The two failure rows belong together: every class depends on the Connect API, so a platform outage reaches everyone regardless of where connector code runs. What the gateway changes is that connector faults become correlated rather than independent — a design input for the gateway (the reference deployment runs a single replica today, so its resilience is durability and recovery rather than replication), and the reason deployment discipline matters more there. On Dynamics, note, there may be no push at all: the scheduled poll is the default and the webhook is opt-in. The connector classes topic works the trade through.

Auto-accept is the row where the connectors disagree most sharply — from a Zendesk setting that accepts everything to a Salesforce flow that will reject a request it cannot verify, and a Fin/Intercom guardrail that keeps the commitment with a person. Accepting stops the SLA clock and tells a partner an engineer is engaged, so which posture you pick is a real decision: deciding what to accept automatically.

Common ground

What Every Connector Does the Same Way

However different they look, all of them converged on the same handful of behaviours — because the API's shape requires it. If you are building your own, these are the parts worth copying rather than rediscovering.

Pattern

The fetch is the truth

Inbound events carry a token and little else, so every connector acknowledges fast and then fetches the case. None of them try to reconstruct state from the event body — which is also what makes redelivery harmless.

Pattern

The token is the primary key

All of them persist the token at creation, before anything else can fail, and key every local record on it. An integration that loses the token has orphaned a case the partner can still see.

Pattern

Push, with a poll behind it

Every one pairs event delivery with a scheduled sweep. Push is the fast path; the poll is what makes missed deliveries recoverable rather than permanent.

Pattern

Forms are fetched, never cached

The partner's process form is retrieved fresh for every submission, because partners change them and the required documentId comes with the fetch.

The behaviours above are described endpoint by endpoint in Working with the Connect API, and specified in TSANET-2026-001.

No connector for your platform

Four Routes, in Order of Effort

Before you start, settle one question: will your platform run your code inside your own tenant? It decides who operates the integration, where credentials live, and whether connector faults are independent or correlated — and it is a property of the platform, not a choice you get to make. The connector classes topic has the four-question procedure for working it out. Building with an AI agent or assistant rather than by hand — on any provider — is a separate question, covered in Agentic usage with TSANet Connect.