Skip to main content
MITHUNAI has two kinds of AI worker, and they do different jobs. An assistant answers questions from your content: it retrieves passages, writes an answer from them, cites them, and says it does not know when they do not cover the question. An agent acts: it plans a goal as steps and calls tools to carry them out. Assistants are the core of the product and are available today. Agents are a separate surface that is in beta, with access on request.

​
Side by side

​
When to use an assistant

  • People ask questions that your documentation, repositories or files already answer.
  • A wrong answer costs more than no answer, so you want citations and an honest “I don’t know”.
  • The audience is public or anonymous: visitors on your website, customers in your product.
Start with Create and configure an assistant.

​
When to use an agent

  • The work is a task, not a question: triaging, drafting a change, running a recurring check.
  • Someone on your team will review what it did, or approve the steps that matter.
  • You are prepared to decide what it may reach.
Agents is in beta, with access on request: read Agents for where it stands before you plan around it.

​
Using both

An agent can consult an assistant. The MCP server gives an agent in a framework you run yourself three read-only tools to list your knowledge, search it and ask an assistant. The answer it gets back is grounded and cited like any other; what the agent does with it is the agent’s own action, outside the answer engine. An assistant cannot do the reverse: it does not call tools or act in other systems, and no guidance you write can make it.
Last modified on November 7, 2026