Yes — Google Gemini connects to SAP Business One through a custom Gemini app that reaches your ERP over MCP. TEKAI provides the connection as a running service, tested and working since August 2026. Approved users then ask SAP Business One questions inside Gemini itself.
Yes — tested and working with TEKAI as of August 2026.
There is no native integration between the two products, so the connection is made over MCP: an open protocol for letting an assistant reach a system it was not built against. The bridge translates the question, applies the asking user’s permissions and records what was asked.
Gemini reaches that bridge through its own app framework rather than through a generic plug-in, which is the one part of the setup that differs from the other assistants. Everything on the SAP Business One side is identical.
Gemini connects through a custom Gemini app: TEKAI is added once via Gemini’s app framework (Spark), and approved users then ask their SAP Business One questions directly inside Gemini.
“Added once” is the operative phrase. The app is registered a single time for the organisation, after which provisioning is a per-user matter rather than a per-user installation.
MCP is model-agnostic, and one TEKAI endpoint serves all six supported assistants — Claude, ChatGPT, Gemini, Microsoft Copilot, Grok and Perplexity. A team that starts on Gemini and later standardises elsewhere keeps the same endpoint and the same permission model.
Stock and production questions are where this connection tends to prove itself, because the answer usually requires joining three tables nobody wants to join by hand.
Dead and slow-moving stock is the standard first question: items with no movement in a given period, their on-hand value by warehouse, and the last date each one sold. That is a genuinely awkward query to build and a trivial one to ask.
From there the same connection covers stock-out risk against open orders and lead times, batch traceability in both directions from a lot number, and yield against raw input for anyone running production orders. Cross-company stock — where the same item sits in two databases — is a common reason to connect an assistant at all.
The pattern worth noticing is that none of these are new capabilities. The data was always there; what was missing was a way to ask without commissioning a report.
Steps two and three are the ones that decide whether the result is useful. The rest is configuration.
Permissions, logging and data handling are the same as for every other assistant: read-only by default, scoped per user, grounded in your data and never trained on it.
Because provisioning is per user, the warehouse supervisor and the finance controller reach different parts of the system. The bridge honours the boundaries your ERP already enforces — which is a reason to tidy loose authorisations before connecting anything, not after.
Write actions, where you want them, are added deliberately and still confirm before posting. Full detail is on the SAP Business One AI agent page.
The work that decides whether the answers are useful happens before anything is connected.
Item groups are the usual weak point. If half your stock sits in a group called “General”, every question about category performance returns a shrug — and the assistant cannot tell you that, because from its side the data looks complete.
Units of measure are the second. Where the same item is bought in cartons, stocked in pieces and sold in kilograms, stock-value questions produce numbers that are individually correct and collectively meaningless until the conversions are right.
Then reorder levels and lead times, which are what turn a stock question into a forecast. A stock-out prediction is only as good as the lead time it is measured against, and lead times in most systems were entered once and never revisited.
The encouraging part is that this work pays for itself whether or not you connect an assistant. Cleaning item groups improves every report you already run. If a project stalls here, it has found a real problem rather than created one.
The practical difference between assistants is the registration step, not the capability. Gemini uses its app framework, Grok uses its connectors, Microsoft Copilot goes through Copilot Studio, and Claude, ChatGPT and Perplexity use the native bridge directly.
Choose on the basis of what your team already has. If the organisation runs on Microsoft, Copilot on SAP Business One is the more natural fit; if you want the assistant SAP itself now builds on, see Claude on SAP Business One.
The six routes and the build-versus-buy arithmetic are compared in the guide to connect AI to SAP Business One.
Yes. It has been tested and working with TEKAI since August 2026. Gemini connects through a custom Gemini app built on Gemini’s app framework, which reaches live SAP Business One data over MCP.
It is how Gemini registers an outside capability. TEKAI is added once as an app, and from then on approved users ask their SAP Business One questions directly inside Gemini rather than switching to another tool.
You need a Gemini subscription that your users can access. The bridge is the same regardless of which assistant is asking, so the requirement sits on the Google side rather than on SAP Business One.
Yes, where those databases are part of your installation and the asking user is permitted to see them. Cross-company stock and cost questions are one of the more common reasons businesses connect an assistant in the first place.
No. Answers are grounded in your data and your data is not used to train the model. Access is read-only by default, provisioned per user, and every question is logged for audit.
Bring the item nobody can explain. Thirty minutes on your live system.