SAGE INTACCT
Upgrade Safe Architecture for AI Agents on Sage Intacct
Sage Intacct ships new releases on a regular cadence, and a fragile integration breaks with each one. The durable pattern keeps AI outside the core, talks to Intacct through supported interfaces, and treats the AI Gateway as the seam. Here is what upgrade safe looks like.
- The durable pattern keeps AI outside the Intacct core and talks through supported interfaces only.
- Sage recommends the REST API for new work, and new objects ship there first, so building on REST reduces future rework.
- The AI Gateway and MCP Server are designed as the seam for external AI, which is exactly where a layer should attach.
- Avoid deep customization of the core to bolt AI in. That is the part most likely to break on a release.
- ChatFin attaches at the supported seam and carries forward across releases without core changes.
Every general ledger that gets updated on a schedule punishes brittle integrations. Sage Intacct ships releases regularly, and an AI layer wired into the wrong place breaks with each one. Sage Intacct exposes two programmatic interfaces: a modern REST API that reached general availability in 2025 and is now the direction Sage recommends for new work, and the older XML API that Sage labels Legacy but still supports. New objects and features ship on REST first, while a handful of long standing objects and workflows still live only in the XML layer.
The good news is that Sage has given AI a proper place to attach. In 2026 Release 2, shipped on 8 May 2026, Sage added the Sage Intacct AI Gateway: a governed way for external AI tools to reach Intacct data through the REST API and a Sage Intacct MCP Server that speaks the Model Context Protocol. Compatible AI clients can explore metadata, query records and generate insights in natural language under controlled access. Building against that seam, rather than into the core, is what makes an AI layer upgrade safe.
What breaks on a release, and why
Integrations break when they depend on things that move: an internal screen, an undocumented behavior, a heavy customization layered into the core. When Sage ships a release, those assumptions shift and the integration fails quietly.
The fix is to depend only on supported, versioned interfaces, and to keep your logic outside the ERP.
Releases are the constant
Sage Intacct ships updates on a fixed rhythm, and each one is a chance for a fragile integration to break. The pace is not slowing down.
The only constant in this industry is change. The pace of that change is extraordinary.
Craig Sullivan, on the AgentCFO podcast
Design the AI setup below on the assumption of constant change. Anything that depends on a screen or field staying put will need fixing again.
Build on REST first
Sage recommends the REST API for client applications, and new objects and features are released on REST going forward. Building on REST rather than the Legacy XML layer means you are aligned with where Sage is investing, which reduces future rework.
The XML API remains supported for the objects that still live only there, so a mixed approach is fine as long as REST is the default.
Treat the AI Gateway as the seam
In 2026 Release 2, shipped on 8 May 2026, Sage added the Sage Intacct AI Gateway: a governed way for external AI tools to reach Intacct data through the REST API and a Sage Intacct MCP Server that speaks the Model Context Protocol. Compatible AI clients can explore metadata, query records and generate insights in natural language under controlled access.
A seam that Sage designed for external AI is exactly where your AI layer should attach. It is meant to be stable, and it keeps Sage in control of access.
Keep logic outside the core
The single biggest cause of upgrade pain is business logic pushed into the ERP as customization. Keep the reasoning, the drafting and the orchestration in an external layer. Let Intacct be the system of record, and let the AI layer be replaceable without touching the core.
Governance that survives change
Scope a dedicated integration role, keep least privilege, separate read and write credentials, and keep a person on every posting. None of that depends on a release, so it does not break when one lands. See security and governance for the full pattern.
How ChatFin stays upgrade safe
ChatFin attaches at the supported seam. It reads through the MCP Server and REST API, writes through the REST API on approval, and holds its own reasoning outside Intacct. When Sage ships a release, the core is untouched, so there is nothing in the ERP to break. ChatFin does not replace Sage Intacct or compete with Sage Copilot. It runs on top of the ERP, and where the AI Gateway, MCP Server and REST API are available it uses them as the governed pipe Sage built. What ChatFin adds is the finance workflow layer: multi step processes from capture to posting, run across every entity, with a human approving each write back and a full prompt to transaction trail.
A test before you commit
Ask any vendor a simple question: what happens to your integration on the next Sage release. If the answer involves reworking a customization inside the core, that is a maintenance bill waiting to arrive. If the answer is nothing, because we attach at the supported interface, that is the pattern you want.
An AI Layer That Survives Every Sage Release
Upgrade safe means keeping AI outside the Intacct core, building on the REST API where Sage is investing, and attaching at the AI Gateway seam designed for exactly this.
ChatFin attaches at the supported interface and carries forward across releases with no core changes. Book a demo to see the architecture.