Writes
Run as the user’s own ERP credentials. The ERP enforces its own write rules. We do not second-guess them.
A guided tour for finance and operations leaders. What the platform is, how it keeps each user’s ERP permissions intact, and what going live actually takes.
Most AI-for-ERP tools connect with one service account and hope nobody notices. That account can read everything, so the AI can too. We took the harder path.
Two people ask the same agent the same question: “show me project profitability.” One is a company accountant with access to a single entity. One is a group financial controller. They get different answers, because their ERP permissions are different.
Run as the user’s own ERP credentials. The ERP enforces its own write rules. We do not second-guess them.
Run as the user’s own ERP credentials. Never a shared account. Tunable per tenant if your policy is stricter than ours.
Service account, filtered to that user. Fast and cacheable. The user’s role rewrites every query before the agent sees a row.
We never re-implement your permission logic. We pass identity through.
Every action an agent can take declares its class, so the runtime picks the right identity path without per-agent wiring.
The permission paths ship with tests that actively try to break them. A leak is a P0.
When your access covers some entities and not others, an aggregate says so and names what it left out. The alternative is quietly dropping rows.
Most mid-market finance teams run more than one entity, and every ERP models that differently. A company switcher sets the active company, and the agent scopes every answer to it.
Company access follows each person’s own Intacct role. One connection covers every company that person can already see.
Each company sits in its own database. Ask for a group figure and we gather it company by company, then tell you which ones your access did and did not cover.
The same per-person company model, with an order posted into a live system to prove the path end to end.
Next on the connector list. The same company model applies when it lands.
Covered today by our advisory and custom builds. A platform connector is on the roadmap.
The permission model travels with the agent, so it holds wherever your team works.
Chat, approvals, admin, audit and the agent catalog at your own subdomain.
Your team’s own AI client calls the same agents. Permissions come along, so there is no client that can suddenly read everything.
Sage shipped Agent Builder in its R2 release. When we plug into it, the agents appear in Sage itself. Microsoft Copilot, Teams and ChatGPT follow the same path.
An open standard that lets AI assistants call business tools safely. Build the agent once and plug it into Claude today, with Copilot and ChatGPT to follow.
On SAP Business One there is no native AI to compete with, so we build the surface ourselves. On Sage Intacct, Sage’s R2 release shipped Agent Builder and a native MCP server, so we plug in and let Sage carry the chat layer. On MYOB Acumatica we lead with advisory and custom builds. Same agents, surfaced where native AI does not already cover them.
It uses your data. It never trains a model on it. Answers are grounded in your live records plus context you approve: your terminology, your reports, your conventions. Retrieved fresh on every question, never baked into a model.
That distinction is the value. Fine-tuning on your data is hard to audit, hard to undo, and weakens the security story. Retrieved context is the opposite: visible, editable, deletable, and it is what makes the agent feel like it already knows your business.
When your team says supplier and the ERP says vendor, or your dimensions are Department by Project, the glossary captures it. Built with you at onboarding, editable by your admin afterwards.
Register an Intacct report or a B1 saved query you have already tuned, and the agent reaches for it when a question matches instead of rebuilding the answer from raw data.
No fine-tuning on customer data. Not now, not planned. Today the agents ground answers in your live ERP data and cite the source records. The glossary and registered reports are how we plan to make that feel native to your business.
Agents prepare the work and show you what they found. A person checks it and approves it. Only then does anything reach your ERP, and every step is recorded.
This path runs end to end on SAP Business One and SAP Cloud ERP: a real order, approved by a person, posted into a live system. Today it is used for sales orders. Other document types follow the same rule: draft, review, approve, post.
nextBringing the same approach to fully on-premise installs.
From contract signed to your team asking real questions. Sequenced, predictable, with a senior Leverage Technologies consultant on the line rather than a self-serve wizard.
Currency, financial year, timezone, GST display, multi-entity, your subdomain. About 30 minutes.
OAuth for Sage Intacct, a guided connection for SAP Business One. We walk your IT lead through it.
A senior advisor sits with your finance lead to capture your terminology, dimensions and conventions.
Your existing custom reports become things the agent can call by name. Today, agents ground answers directly in live ERP data.
Users sign in with their own accounts. ERP consent happens once per user, on first use. You are live.
The terminology conversation can be a short admin walkthrough, a guided first cut with an advisor, or an entity-by-entity pass across hundreds of GL accounts. We size it to what will make the agents feel native.
Contract to connection in about a day. Terminology alignment runs one to two weeks of advisory time in parallel.
Everything is done on screen: adding a person, connecting your ERP, changing what someone is allowed to see. You do not need a developer to keep it running.
Hosted in Sydney. Your data sits in its own space in a standard database, your ERP credentials are encrypted, and every action is logged. Nothing unusual, nothing your IT team has not reviewed before. The detail is below.
For a deeper architectural picture, internal review only, talk to your Leverage Technologies contact.
Start with an AI Value Workshop, or join the waitlist for pilot access.