How EdgeKeeper connects AI to your business tools

An orchestration layer is the software between an AI model and your business systems that decides what the AI can reach and do: which tools it may use, which records it may read for the person asking, which instructions it follows, and which actions must wait for a person to approve them. The model writes the answer; the orchestration layer controls what it works with.

For a business owner, its value is practical: the assistant can work with the right context while access and responsibility remain clear. This guide explains what the layer does, how AI agents rely on it, and what to ask before you trust one with your systems.

Start with one useful workflow

Consider preparing a quote. The work may involve customer records, past jobs, a price list, and a document template. A useful system brings the approved information together, produces a draft, and puts it in front of the person responsible. It should show what information it used and stop when it cannot complete the task reliably.

Connections, context, and controls

• Connections: agreed tools that make selected information available.
• Context: relevant records and business procedures, scoped to the person asking.
• Instructions: how your team wants the task performed and checked.
• Controls: logged access, individual approval for changes, and an escalation path.

Stored business records are different from the information sent to a model. EdgeKeeper uses a dedicated processing server and application records scoped to your business. The security disclosure explains those boundaries and the assurance we do and do not hold.

AI agents explained: how agents use the orchestration layer

What is an AI agent? It is an AI model given a goal and a set of tools. Instead of answering one question, it works through steps: find the customer, check the last three invoices, look up the price list, draft the follow-up. Each step is a request to a connected system, and every request passes through the orchestration layer.

That is where the limits live. The layer decides which systems an agent can reach, which records it may read for the person who asked, and whether it can change anything at all. Most agents should be read-only: they look things up, summarise and draft, and a person decides what happens next. In Switchboard, your team can create read-only agents just by asking.

An agent that does make changes, such as updating a job status or sending an appointment reminder, is built with you and scoped to that one task. Each change it proposes waits for a person to approve it, and the request, the records it used and the decision are logged. When an agent meets something it cannot do reliably, it should stop and hand the work back rather than guess.

The glossary defines read-only agents, human approval and prompt injection in more detail.

AI agents vs chatbots

A chatbot answers what you type, using what you paste in. It does not look anything up in your systems, and when the conversation ends, nothing has happened in your business. An agent works toward a goal across several steps and uses connected tools to do it. That makes it more useful and also more consequential, because each step reads or touches real records.

The practical difference is accountability. With a chatbot, the person copying the answer is responsible for what happens next. With an agent, the orchestration layer has to carry that responsibility: it limits what the agent can reach, records what it did, and holds any change until a person approves it.

Orchestration layer vs workflow automation tools

Workflow automation tools move data between apps when a trigger fires, following steps someone defined in advance: when a form arrives, create a contact and send an email. They are reliable for fixed, predictable sequences, and many businesses already use them.

An orchestration layer covers the work that does not follow a fixed path, where an AI model reads the situation and chooses the next step. That needs things a fixed sequence does not: per-person access to records, a log of what the model read, limits on which tools it may call, and a point where it stops and asks. The two often work together. A fixed automation can start a job, and the orchestration layer governs the part where AI is involved.

Questions to ask about any orchestration layer

• Reach: which systems can it connect to, and who agreed each connection?
• Scope: does an agent see only the records the person asking is allowed to see?
• Changes: which actions are read-only, which can write, and who approves a write?
• Records: can you see later what was asked, which records were used and what was decided?
• Failure: what happens when a connection is down or the model is unsure? It should stop and hand back, not guess.
• Ownership: can you export your data and the business knowledge it has collected if you leave?

Where EdgeKeeper fits

EdgeKeeper is our proprietary orchestration layer, built by IT and security pros and provided on a Microsoft Azure server dedicated to your business. We configure connections and build the automations in your proposal, then manage and support them. Your staff do not need to maintain a server in the office.

External services can be connected when the agreed work requires them. A connection does not grant unlimited permission: its data access and actions must fit the documented controls.

On top of that foundation, your team works in Switchboard. It gets answers, reports and dashboards from the connected systems, and can make as many read-only AI agents as it wants, just by asking. Anything that makes changes is built with you and approved by a person.

Keeping it useful

Business processes, integrations, and models change. New versions need validation; improvements are not automatic. We review failures, maintain the work we delivered, and agree the scope of additions. Your data and accumulated business knowledge remain yours to export.

The first step is measuring the workflow with Edgerton. The report and proposal let you decide what is worth building before committing to EdgeKeeper and ongoing support.