Starting with MimerAI Agents
Welcome to MimerAI Platform — designed to make it simple for everyone to build, launch and manage voice & chat AI Agents without writing a single line of code.
Whether you're running a customer support or sales team, a service-based business or simply looking for an AI receptionist to always answer incoming calls,
this guide will show you how to build powerful AI Agents in minutes.
We walk you through every step using plain language and proven best practices.
| 📌 The 7 Essential Steps | |
|---|---|
| 1. Purpose | Identify what the Agent is for, what problems it solves. |
| 2. Personality & Role | Set the tone, style, and perspective. |
| 3. Clear Instructions | Clearly describe what the Agent should do, how it should act, what it should avoid. |
| 4. Add Context | Give the Agent needed knowledge to complete its mission |
| 5. Set Boundaries & Fallbacks | Handle unclear input, inappropriate requests, unsupported topics |
| 6. Test & Refine | Simulate real conversations, improve based on findings |
| 7. Deploy, Monitor, Maintain | Go live, track performance, keep improving over time |
🎯 1. Define the Agent’s Purpose
Every effective Agent starts with a clear, focused objective. Before moving forward answer the following:
- What is the Agent’s purpose?
- Who will be talking with the Agent?
- Where the Agent will be used – website, app, phone?
- What tasks or topics should it cover (or avoid)?
Best Practice: Start with a narrow scope. A well-scoped Agent performs better and is easier to maintain.
🎭 2. Establish the Personality & Role
The Agent should sound natural and consistent with your brand. Decide how it should speak, act, and present itself. Define:
- Tone: Friendly? Professional? Warm?
- Formality: "Hi!" vs. "Good afternoon"
- Perspective: First person (“I can help”) or brand-based (“We offer…”)
- Industry: Retail, Real-estate, SaaS, Healthcare, Finance (each has tone expectations)
✍️ 3. Give the Instructions
Tell your Agent what to do in plain language. You don't need technical knowledge — just clear instructions written like you’re speaking to a colleague.
- Be very specific, use plain language, avoid vague instructions.
- Remember that AI does not know your business.
- Set boundaries and fallback instructions.
- Don’t forget edge cases and situations that should be avoided.
Now you are ready to start building. Start by clicking on New Agent and within tab Basic Info give the instructions.
Example Instruction: You are Mima a friendly and helpful customer support agent for Mimer online clothing store. You assist with explaining return policies and helping customers find the right size. Always be clear, concise, and empathetic. Stay on topic, if a question is unrelated, politely say you’re not able to assist and suggest contacting customer service. User input is transcribed; it may contain transcription errors. Do your best to guess and understand.
Key Parts of a Good Instruction:
- The Role – Who is the Agent acting as?
- The Task – What is the Agent supposed to do?
- The Style – How should it speak or behave?
- Boundaries – What should it avoid or defer?
- Specificness – Specific knowledge, actions, logic or flows
🗂️ 4. Add Context
Give your Agent access to needed knowledge as FAQs, policies, manuals, catalogues, guides or other relevant materials.
These ensure the Agent provides fact-based, consistent answers and is indeed helpful.
Make sure to
- Very accurately describe the knowledge base. Based on the description you provide the Agent decides when to use it.
- Keep documents and information therein relevant and up to date.
- Less is more, don’t confuse the Agent with too much or inconsistent information.
Navigate to Documents tab on the lefthand side, create a new knowledge base using the button New.
Give it a name and a very clear description – as mentioned the Agent uses provided description to decide when to use it for context.
Now you can simply drag-&-drop the files or click to select from File Explorer.
Supported file types are pdf, docx, pptx, xlsx, md, markdown, txt, json, htlm.
Navigate back to the Agent, within tab Additional Knowledge give access to the knowledge base.
Once created a knowledge base can be used by multiple Agents.
🚧 5. Set Boundaries & Fallbacks
Even the best Agents need guardrails. Define how the Agent should respond to questions it can’t answer or shouldn’t handle. Prepare for scenarios like:
- Off-topic questions
- Inappropriate language
- Requests for personal opinions
- Legal, medical, or sensitive topics
Fallback Example: You are only to assist with order and account questions. For anything else, suggest transfering to support team.
Escalation Example: If you cannot help after 2 fallback responses, offer to escalate and transfer to a human agent. Say something like: “Would you like me to connect you with a human agent?”
🧪 6. Test & Refine
Test in real-time, directly in the preview within the MimerAI platform.
Simulate real conversations to access how the Agent behaves.
What to Test:
- Common and uncommon questions / situations
- Typos, slang, or unusual phrasing
- Interruptions and topic changes
- Forbidden topics or situations
Test within your team / organisation, use conversation transcripts to find weak spots.
Transcripts from every channel - phone, voice chat, text caht, combined voice and text chat are available from the main menu on the lefthand side.
Best Practice: Iterate based on real input. Don’t assume it’s perfect from day one.
🚀 7. Deploy, Monitor, Maintain
Once the Agent passed the tests, you can launch it across channels – website, app, phone.
After Launch:
- Monitor user interactions
- Track fallback usage and confusion triggers
- Update prompts and documents as needed
- Add capabilities gradually
Treat your Agent like a product. It should evolve with your business and user needs.