Agents that work inside your systems, not beside them.
TL;DR. Business Agents act over ERP, CRM, M365, ticketing and your own APIs — with tool-use, sandboxing and per-tenant permissions. Every action is governed by the LLM Gateway, bounded by role, and logged for audit. They do the routine; humans approve what matters.
What this is about
A chatbot that can only talk is a toy. A business agent that can act — open a ticket, update a record, reconcile an invoice, draft and file a document — is useful, and dangerous if ungoverned. Business Agents give you the action with the guardrails: scoped tool-use, explicit permissions, and an audit trail for every move.
How we run it
Agents connect to your systems through defined tools (ERP, CRM, M365, ServiceNow, custom APIs), each with explicit permissions per tenant. The LLM Gateway governs the model calls; the Message Bus coordinates multi-step work; Observability records every tool call and decision. Consequential actions pass a human approval gate. Sandboxing keeps an agent inside its remit — it sees and touches only what policy allows.
When it fits
Organisations with high-volume, rule-heavy back-office work — order handling, invoice processing, service tickets, onboarding. Teams that want automation that acts in real systems, not just suggestions. Regulated businesses that need "who did this, and why" answerable for automated actions.
What we don't do
We don't give agents unbounded access — every tool and permission is explicit. We don't hide the logic in the model — actions trace to rules and are logged. We don't rip and replace your systems — we wire agents into the ones you run.
What you can hand off
-
System integrations
Defined tools over ERP, CRM, M365, ServiceNow and your custom APIs.
-
Scoped permissions
Per-tenant, per-tool permissions — agents touch only what policy allows.
-
Human approval gates
Consequential actions require sign-off; routine actions run and are logged.
-
Sandboxing
Agents bounded to their remit; revoke immediately if behaviour drifts.
-
Full audit trail
Every tool call, decision and approval recorded — provable after the fact.
Product facts
| Integrations | ERP, CRM, M365, ServiceNow, custom APIs |
|---|---|
| Governance | Model calls via the LLM Gateway · actions via policy |
| Permissions | Per-tenant, per-tool · explicit, revocable |
| Approval | Human gate on consequential actions |
| Audit | Every tool call and decision logged |
| Deployment | EU cloud, on-premise or air-gapped |
Asked before the briefing
-
Can an agent do something destructive on its own?
No. Consequential actions pass a human approval gate. Routine, low-risk actions run automatically and are logged. -
How do agents reach our systems?
Through defined tools with explicit, per-tenant permissions. An agent sees and touches only what policy allows. -
Do we have to replace our ERP/CRM?
No. Agents integrate with the systems you run — ERP, CRM, M365, ServiceNow, custom APIs. We orchestrate, not rip-and-replace. -
Can we prove what an agent did?
Yes. Every tool call, decision and approval is recorded. 'Who did this and why' is answerable.
Automation that acts, with the guardrails on.
We show business agents acting in a sandboxed copy of your stack — scoped permissions, approval gates and a full audit trail.