Combined vs Paragon for AI agent data access

Compare Combined with Paragon ActionKit and Managed Sync: query-layer ownership, customer identity, permissions and a Salesforce–Zendesk evaluation.

By the Combined team · September 10, 2026

Choose Combined when you want connected business records and managed SQL for an existing agent. Choose Paragon when you are building integrations into your own product, including customer onboarding, application tools and data replication. Paragon supports both live operations and Managed Sync; the difference is how much of the downstream application and query layer you want to own.

A business answer or an integration product

“Which valuable accounts have unresolved support work?” is a useful first project for Combined. Connect the CRM and support sources, grant your agent access and calculate the answer across managed datasets. You can keep the agent you already use without building a separate customer-facing integration interface or query service for that job.

Paragon fits a different starting point: your product needs customers to connect their applications, then use those connections inside the product. Its ActionKit supplies tools and triggers, while Managed Sync supplies ingestion for downstream applications, including analytical use cases. Compare the responsibility split before comparing connector lists.

From source connection to an analytical answer
Responsibility	Combined	Paragon
Connection experience	Connect sources in your Combined workspace.	Embed customer connections through the SDK and Connect Portal.
Data ingestion	Maintain selected source datasets in managed storage.	Managed Sync normalizes supported records and exposes them through APIs.
Analytical query path	Discover logical datasets and run managed read-only SQL.	Pull synced records into your database or retrieval pipeline and implement the query path.
Application operations	Read-only data surface for the agent.	ActionKit tools and triggers support reads, writes and event-driven features.
Business definitions	Your team defines account mappings, measures and windows.	Your team defines account mappings, measures and windows.

Paragon's production Sync API pattern describes the downstream database handoff. Combined's SQL interface is the managed calculation surface. That distinction matters when the goal is to get a repeatable answer into an agent's workflow.

Paragon already handles substantial integration work

Managed Sync handles ingestion, normalization and recovery from provider problems. Its model includes CRM contacts, companies, deals and custom objects, plus ticket records. Unmapped CRM fields remain available in customFields; ticket source fields are available in originalRecord. See Paragon's Synced Objects.

Your application consumes that data, maintains its destination, and implements the analytical access it wants to expose. With Combined, those datasets and the read-only query service are already part of the product. Your agent discovers actual tables, checks freshness and asks the next SQL question. This is the focused choice when you want the data connection to be the finished deliverable.

Check freshness across the complete path. Paragon documents minute-level incremental checks and periodic full refreshes; destination processing still matters. Combined supports manual, daily and account-gated hourly schedules. Match successful commits to your decision's needs using the respective sync lifecycle and Source state.

Keep customer identity, credentials and reporting accounts separate

In Paragon, a Connected User is often a customer organization, but your application can choose another entity. The JWT subject identifies it. By default, its token can access that Connected User's credentials and configurations; JWT Permissions can narrow that scope. These are important application design decisions. See Connected Users and JWT Permissions.

Managed Sync's Permissions API adds source-derived object checks for supported integrations. Its current documented coverage is a selected set of file and document systems, and the application applies those checks in retrieval. Do not assume the same object-permission coverage for every CRM or ticket pipeline. See the supported scope and implementation.

Combined uses explicit Account, Source and optional dataset grants. A provider's production and sandbox tenants remain separate Sources, and ordinary grants do not expand to future connections. Its access model controls which datasets an agent can read.

The reporting map is another layer: which Zendesk organization belongs to which Salesforce account? Normalizing both systems into consistent schemas does not settle that business relationship. Keep a maintained ID map, flag conflicting matches and decide how subsidiaries roll up before calculating value.

Evaluate Salesforce account value against Zendesk backlog

Use our Salesforce–Zendesk worked guide, SQL fixture and expected result. The synthetic example counts unresolved support tickets for accounts with at least $10,000 in USD closed-won opportunity value from June 12 inclusive to September 10 exclusive, 2026 UTC.

Atlas should show $20,000 booked value, two unresolved tickets and one high-or-urgent ticket. Birch should show $15,000 and zero unresolved tickets. An older pending ticket still belongs to current backlog; a solved urgent ticket does not. Closed-won value is a bookings proxy, not recognized revenue or cash.

In Combined, connect Salesforce and Zendesk, verify the first syncs, grant access and discover the required fields.
In Paragon, configure the relevant syncs and consume the records into your application. Its Salesforce pipeline maps Accounts to Companies and Opportunities to Deals; Zendesk supports ticket sync.
In either implementation, aggregate qualifying opportunity value and distinct ticket counts independently, then join through the maintained account map.
Check that a complete zero-backlog account survives. Treat missing source coverage or an absent map separately from zero.

Paragon documents the required pipelines for both Salesforce and Zendesk. Adapt discovered schemas to the fixture's teaching aliases; do not assume they are the actual relation names in either platform.

Using my granted CRM and support sources, find accounts with at least
$10,000 of USD closed-won value in the agreed reporting window.
Add distinct currently unresolved tickets and high-or-urgent counts.
Discover schemas, validate the organization-to-account map and check
freshness first. Aggregate opportunities and tickets separately.
Return definitions, SQL, mapping exceptions and supporting records.

Download the platform evaluation worksheet to record the answer and who owns the destination, updates, permissions and query runtime. The fixture keeps the calculation fixed while you evaluate the full application path.

Common evaluation questions

Can Paragon connect to an agent through MCP?

Yes. Its official MCP implementation exposes ActionKit operations and supports connected-user authorization. Combined exposes a managed MCP endpoint for dataset discovery, freshness and read-only SQL. Compare the tools your workflow needs.

Can Paragon retrieve data as well as change records?

Yes. ActionKit includes retrieval operations, and Managed Sync supplies replicated data. Choose Combined when you also want the managed analytical query layer ready for your agent.

Which cost should a small team compare?

Compare the whole proposed setup. Paragon uses Connected Users as a subscription usage boundary; Combined measures monthly active records. Include the database, query service and model costs your design needs. The units alone do not establish a cheaper option.

Start with the answer your team needs

Open the Salesforce and Zendesk plan for a ready-to-copy assignment and setup references. If your goal is useful business analysis in your existing agent, Combined brings the ingestion, data and query service together. Start with an initial 5 million MAR without a card and connect your first sources.

Sources and further reading

Explore the documentation behind this guide. Product details checked on September 10, 2026.