Service agent · 1 row · account holder onlyCredit agent · needs approvalWaiting · Karen, credit managerWho asked · tool · permission · approverWho asked, which tool, which permission, who approvedEach action your system allows, published with a plain name and description: “get account balance”, “create sales order”, “update ticket status”.
Per agent and per tool: read only, read and update, update with a person’s approval, or no access.
The number comes from your ERP or CRM on every call, not from the AI’s memory.
A record of who asked, which tool ran, on what data and who approved it.
get_account_balanceCustomer“What do I owe right now?”
1 row · account holder only
Agent“Your balance is $1,860.00 across 2 open invoices.”
sales_by_branchManager“How did each branch do this week?”
3 rows · no personal data
Agent“Dallas leads the week, Phoenix is close behind.”
change_credit_limitCustomer“Can you raise my credit limit to $5,000?”
Needs a person's approval
Waiting · Karen, credit manager
Approved by Karen
Agent“Approved. Your credit limit is now $5,000.”
Every call logged: who asked, which tool, which permission, who approved.
| Step | What our team does | What you approve |
|---|---|---|
| 1. Documentation | Reads your API documentation (an OpenAPI file, a collection export or a PDF) and detects the available actions | |
| 2. Tool list | Proposes which actions to publish, each with a plain name | Which tools go live |
| 3. Credential | Receives a credential from you through a secure channel; it is stored encrypted and never shown again | |
| 4. Live test | Runs a real call and shows you what each tool returns | That the data is right |
| 5. Permissions | Sets what each agent may do, tool by tool | Who sees and does what |
| 6. Ready | Your agents in Konnet Flow start using the server |
When a customer asks for a higher credit limit, check their balance and payment history in our ERP, and send any increase to the credit manager for approval.
Done: 6 steps. 1 waits for a person's approval and the rest run on their own.
| Step | What the agent does | What a person approves |
|---|---|---|
| 1. A buyer writes on WhatsApp “Can you raise our limit to $5,000?” | Understands the request with a low-cost AI model | |
| 2. Checks the account | Calls get_account_balance and payment_history, by direct code, no tokens | |
| 3. Applies your rule | Any increase goes to credit, by policy | |
| 4. Prepares the case | Puts balance, history and the amount requested in one view | The credit manager approves, adjusts or declines |
| 5. Updates your ERP | Calls change_credit_limit with the approved amount | |
| 6. Replies and logs | Confirms to the buyer; the call and the approval land in the audit trail |
This flow is built in Konnet Flow: you describe the process, the Copilot builds it and you approve it before it goes live.
See Konnet FlowEvery tool is validated field by field against your source. A field that doesn’t exist is rejected before any agent sees it.
Personal data stays out of technical logs and out of answers to third parties unless you authorize it.
Credentials are encrypted and never shown again; each company’s data is isolated.
Ask us to pull an agent’s access and we do it without touching the rest.
When Flow Copilot proposes a change, it shows the plan and waits for a person to approve it.
For companies in New York, Los Angeles, Chicago, Houston, Miami and across the United States: setup and go-live happen remotely.
“How much did we sell this week by branch?” or “Which accounts are more than one cycle past due?” The agent answers on WhatsApp or web chat with the figure from your system.
A service agent sees the balance of the customer who is writing; an analysis agent sees sales but can’t create orders.
AI-generated dashboards are a capability our team already uses internally and is preparing for clients.
Konnet publishes your systems as a standard MCP server (JSON-RPC) for your agents in Konnet Flow.
If your team wants to use the same tools from agents you build yourselves, or through a REST API, we enable it on request.
Konnet Flow can also use third-party MCP servers as one more source, with the same permissions and logging.
Each new source is configuration on the same platform, not a development project.
An MCP server is the doorway an AI agent uses to work with one of your systems. MCP stands for Model Context Protocol, an open standard that defines how an agent discovers the actions available, calls them and gets the data back. For a business, it means your ERP, CRM or billing system can be used by AI agents without a custom integration for each one. With Konnet, our team sets up the server and your agents in Konnet Flow use it with the permissions you grant.
An API is how your system takes orders from other software, and someone has to read its documentation and code each call. An MCP server sits on top of that API and presents it as a list of tools, each with a name, a description and a permission, that an AI agent understands and uses without custom code. It isn’t one or the other: our team reads your API documentation with Konnet Core, our internal tool, and publishes the MCP server on top of it.
With a documented API, our team has the connection ready in about 5 minutes. You share the link or file with your API documentation, we detect the available actions and propose which ones to publish, you send a credential through a secure channel, and a live test confirms the data comes back right. Then you approve what each agent may do. You don’t write code. Konnet needs an API to connect a system; if yours has none, we tell you before you commit.
Yes, and that control is the core of how Konnet sets up your MCP server. Before an agent goes live, you decide tool by tool what it may do: read only, read and update, update with a person’s approval, or nothing. A service agent can see the balance of the customer who is writing but not your sales; an analysis agent can see sales but can’t create orders. If you ask, we revoke an agent’s access at any time without affecting the others.
It is when every action has an owner and a record, and that is how Konnet works. Your system’s credentials are stored encrypted and never shown again, each company’s data is isolated, and each agent uses only the tools you authorized. Personal data stays out of technical logs and out of answers to third parties unless you allow it. Every call is logged with who, what, when and on which data. Konnet helps you meet your obligations; review the setup with your counsel.
The number doesn’t come from the AI’s memory; it comes from your system. When someone asks for a balance or this week’s sales, the agent calls the authorized tool and your system answers by direct code, on every request. Each tool is validated field by field against the real catalog of your source, so a field that doesn’t exist is rejected before an agent can use it. The AI only understands the question and writes the reply around that data, and every call is logged.
Konnet publishes your systems as a standard MCP server for your agents in Konnet Flow. If your technical team wants to use the same tools from agents you build yourselves, or through a REST API, we enable it on request, with the same permissions and audit trail. Konnet Flow can also use third-party MCP servers as one more source. Our team configures each connection, so you get one set of tools with one set of rules, whichever agent is calling.
Our team sets up the connection as part of your Konnet Flow implementation, which is quoted as a fixed proposal after the demo. After that, Konnet charges USD 0.25 per resolved action, with plans from USD 100 a month. AI spending stays low because checking a balance, reading weekly sales or logging a payment runs as direct code against your system, without tokens. The AI works only where it adds value, with low-cost models and narrow tasks.