How is security and access handled?
Access is least-privilege and scoped per agent. Each agent runs under its own service account with an explicit allow-list of the systems it may use, read-only wherever possible. Every action is logged, and the rules agents follow are kept as versioned configuration, so every change is reviewable, auditable and reversible.
Does it integrate with our corporate directory?
It can. Agents-C can integrate with your corporate directory, such as Microsoft Entra ID (formerly Azure AD), for single sign-on and group-based permissions. Agents and the knowledge base then respect the same groups and access rights your people already have, and removing someone from the directory removes their access too.
Where does it run?
Where it suits you. Agents and apps run as containers under service accounts, on shared infrastructure or on your own servers, never on a personal laptop or a personal login. They are scheduled, logged and restartable jobs, so they are production-grade from day one and keep running whoever is away.
Is our data used to train models?
No. Your data stays yours and is never used to train models. Agents use retrieval-augmented generation (RAG): they search a permissioned knowledge base built over your documents, ERP, data warehouse and other company systems, use only the passages the requester is allowed to see, and cite the source of every answer so it can be audited.
Which AI models do you use?
We build on Anthropic's Claude, and on Microsoft 365 Copilot where you already use it, on business plans whose terms keep your data out of model training. Training for your team covers how each one handles data, including the difference between not training on data and not retaining it.
How do you prevent wrong or made-up answers?
Answers are grounded in retrieval from your own knowledge base, and every answer must cite its source. Anything the agent is unsure about is routed to a person as an exception rather than pushed through, and every action is logged, so a wrong answer can be traced back to the source or rule that caused it.
Which systems does it connect to?
The systems you already run. On live engagements that includes Oracle, a data warehouse, Microsoft 365 (Outlook and shared mailboxes, SharePoint, Excel), Power Automate and market-data services, out of connectors to more than 200 systems. Connectors start read-only: permission to send or write is a separate approval, granted per account. Legacy screens without an API are handled with RPA or Power Automate behind a portal allow-list.
Who owns it afterwards?
You do. Agent rules and connector settings live as versioned configuration rather than in someone's head, every action is logged, and the setup is documented so your own team can run it. The security model is built to survive a person leaving.