Assistants
What assistants are in meinGPT, when to choose an assistant over a workflow, and how to create one step by step.
AI assistants can support you when you frequently work on tasks with similar context. In the Assistants section, you can use existing assistants, edit them, share them, or create new ones. Under the Templates tab, you'll find pre-built assistants organized by department (HR, Operations, Procurement, Marketing, Sales) as a starting point.
If you're dealing with repetitive tasks that don't share similar context, a Workflow is the right solution for you.
Assistant or Workflow?
Both tools in meinGPT solve different problems. This table helps you decide:
| Criterion | Assistant | Workflow |
|---|---|---|
| Flow | Flexible, the model decides based on the request | Fixed steps, defined in advance |
| Tools | The assistant calls tools autonomously | Configured per step |
| Variables | No formal input, free-form conversation | Typed variables (short text, dropdown, file upload) |
| Models | One model per assistant | Different models per step possible |
| File output | Artifacts (Markdown) plus skills for custom formats | Document Output toggle for PowerPoint, Word and Excel |
| Skills/MCP | Yes | No |
| Typical use | Variable, interactive tasks | Recurring processes with fixed structure |
Rule of thumb: If the flow is the same every time and inputs are structured, use a Workflow. If users should ask freely and the model should pick tools on its own, use an Assistant.
Creating AI Assistants
You have two options for creating AI assistants – either choose a template that fits your use case, or create a completely new assistant.
With Template
Under Templates, you'll find several pre-made templates. Browse through them and select a suitable one. Click on its + Create Button to create the AI assistant. The assistant's editing window will open in the next step – customize it to your requirements.
Without Template
If you can't find a suitable template for your task, you can create a new AI assistant from scratch. This way, you can tailor your assistant from the ground up to your specific requirements.
Step-by-Step Guide
A detailed explanation of every editor section is on the Editor page. Use it to look up the components mentioned in this step-by-step guide.
If you chose a template, the required fields are already filled in.
For steps 1–4, you only need to check if the pre-selection is appropriate and adjust as needed.
Describe the assistant
Set an avatar and a name for your AI assistant and describe its task and function in the description field as precisely as possible, so you can quickly identify it later in the My Assistants overview.
Your favorites list only shows the avatar of your assistant, so choose a suitable avatar to have a clear distinguishing feature.
If you want to share this assistant, the name and description should also help other users who haven't worked with your assistant yet understand what task it's suited for and what methods it uses.
Choose a model
Define which model your AI assistant should use. You can choose between a specific model and an automatic model selection that finds the best model for you.
Set the system prompt
Write the system prompt for your AI assistant. Thoroughly describe what task the assistant should handle and with what methods. Best practices and the system prompt hierarchy are documented under Editor → Instructions.
Define conversation starters
For quicker use of the assistant, you can set one or more conversation starters in the Advanced Options section. They are displayed when you open the assistant – select them with a click instead of writing a start message each time.
Select tools
Does your assistant need specific tools to handle its task?
The Tools section is split into five areas: Basic tools (e.g. calculator, charts, memory), Connectors (Live) for external services like Microsoft 365 or Jira, Custom MCP for your own MCP servers, Databases for direct DB access, and Request integration if a tool you need is missing. More on the Editor → Tools page.
Add skills
Does your assistant need to execute code or call external APIs?
Upload pre-built skill packages (.zip files) or click Create Skill to build one interactively via chat. Skills give your assistant the ability to run custom code in a sandbox. More on the dedicated Skills page.
Connect knowledge
Does your assistant need additional knowledge to handle its task?
Connect a data pool or upload documents to provide the assistant with the necessary context. More on the Editor → Knowledge page.
Save
Save your assistant to start using it. The Assistant Check analysis runs automatically on save. You can then select the assistant under My Assistants or start a chat from the preview – you'll find the preview in the top right.
Where to go next
Editor
Detailed walkthrough of every editor section: General, Instructions, Tools, Knowledge, Advanced Options and Assistant Check
Usage
Starting a chat, favoriting, sharing, importing, pinning to the sidebar, multi-language support and API access
Skills
Reusable code packages for deeper integrations — upload skill packages or build them via chat