Question

How to ensure that AI does not suggest a workaround or unsafe solution and takes into account the product, rights, and client context?

Answer

To reduce the risk of erroneous and undesirable responses, Swarmica uses restrictions on the sources and behavior of AI.

The regular AI agent on the channel responds to a specific public question from the client — a public comment in a ticket. It does not have access to internal comments and confidential data protected by confidentiality. It does not check other metadata — the fields of the application, the company, and other properties. For the response, it uses only the published articles from the knowledge base. If there is no solution in the KB, the agent reports this.

The agent has internal restrictions set to reduce the likelihood of fabricated responses and undesirable information. Its behavior can be further customized by editing the prompt.

An AI agent via MCP can take into account the fields and data of Swarmica available to it and allowed by the configuration of its skills. Access to operations is also determined by the rights of the used API token. Such an agent allows you to customize working with the client's context for a specific scenario.

These mechanisms reduce the risk of errors; it is impossible to give an absolute guarantee of error-free responses from generative AI.

Details: Connecting Swarmica functions to LLM via MCP · Capabilities of AI providers