Editor

Your AI assistant is structured into several sections in the editor: General, Instructions, Tools, Knowledge and Advanced Options. These sections define who the assistant is, how it behaves, what it can do, and what knowledge it has access to. While editing, the Assistant Check in the right sidebar runs alongside.

General

The General section holds the core information about your assistant — avatar, name, model and description. These fields help you and your team quickly recognize the assistant in lists, and they give the assistant its basic identity.

Avatar
The avatar is your assistant's profile picture. In the sidebar and in your favorites, only the avatar is shown, so pick something that's easy to distinguish at a glance. Click the avatar to set its background color and symbol (emoji or icon).

Name
Your assistant's name should be short and concise. Pick a name that makes the task immediately recognizable.

Model

The model selection determines which AI model is used in the chat with the assistant. By default Best Model is selected — meinGPT then automatically picks the right model for each request. You can also pin a specific model if you prefer one.

Description
The description should explain the task and function of the assistant — in more detail than the name. Make sure it's understandable to colleagues who haven't worked with the assistant before.

Instructions

The Instructions section is your main lever for the assistant's behavior. Here you write the system prompt.

System prompt
The system prompt is the heart of your assistant – it defines its behavior, what tasks it solves with what methods, and how it responds. A good system prompt contains:

• The persona of the assistant with a description of its role
• The task and how it should be solved
• Additional context for the task
• The desired response format

You can find example system prompts in the Templates.

Best practice — structure with ## sections:

A prompt with clearly separated sections produces noticeably better results than free-running text. This pattern works well:

## Role
You are an expert in contract analysis.

## Task
Analyze uploaded contracts and identify risks.

## Rules
- Always answer in English
- Use the data pool for reference documents
- Use the Charts tool for visualizations

System prompts are merged in this order:

1. Global prompt at workspace level, maintained by the admin (tone of voice, company-wide guidelines)
2. Assistant prompt from this configuration (the role and task of this specific assistant)
3. User input in the chat (the actual request)

Keep each layer focused: the global prompt stays general, the assistant prompt describes the role, the user phrases the request. Duplication across layers leads to fuzzy answers.

Improve button and Tips
In the Instructions area you'll find the Improve button in the top right, which automatically optimizes your system prompt (structure, wording, completeness). Next to it, Tips opens context-specific guidance: prompt tips for the instructions, tool tips for the tools, and knowledge tips for the knowledge binding.

Tools

In the Tools section, you give your assistant capabilities that go beyond plain answering — calling external services, querying databases, executing code. Tools are organized into five areas.

Basic tools
Built-in tools that meinGPT ships with: calculator, charts, memory, code sandbox, web search, translator, image generator and more. These are ready to use without external configuration.

Connectors (Live)

Connectors integrate with external services via live API — Microsoft 365 (Outlook, SharePoint, Teams), Google Workspace, Jira, Confluence, HubSpot, Salesforce, and more. They provide real-time data and can perform actions (e.g. send an email or create a ticket). Click the gear icon to set permissions and default values (such as a default folder) per connector.

Custom MCP

In Custom MCP you connect your own MCP servers to the assistant. This is the right path when you need to integrate internal systems or specialized APIs that don't have a native connector.

Databases

In the Databases area, your assistant gets direct read access to structured data sources — relational databases (PostgreSQL, MySQL, etc.) or integrated tables. Useful for reporting assistants and data-driven analysis.

Request integration

Missing a tool you need? In the Request integration tab you can let us know which connector or API integration you'd like to see. We evaluate requests based on demand and frequency.

Retry behavior
If a single tool call fails, the assistant retries it automatically. A red marker on an individual tool call in the chat view is therefore normal and does not mean the overall response failed. How thoroughly the assistant explores tool calls is controlled by the Assistant mode in the Advanced Options.

Knowledge

In the Knowledge section you give the assistant access to static company knowledge — through two paths:

Upload files
Provide individual documents to the assistant. Drag the files into the upload area or click to browse. Suitable for small, assistant-specific knowledge sets.

Connect a data pool (RAG search index)
Data pools provide larger volumes of company knowledge through a pre-built search index, maintained by admins. Unlike connectors, data pools don't deliver live data but search over a curated knowledge base. Pick from the available pools instead of uploading many individual documents one by one.

Advanced Options

In Advanced Options you configure conversation starters and the assistant mode — both levers that shape the user experience and the assistant's tool behavior.

Conversation starters
Conversation starters are predefined opening messages, shown when the assistant is opened. Set a starter for the most common use cases (for example "What is Apple's current stock price?" for a finance assistant), so you and your team don't have to type the same opening message every time. Use Add another to add as many starters as you need.

Assistant mode
The assistant mode controls how thoroughly the assistant explores problems with tools — that is, how many tool calls it may make per response. By default Assistant: up to 5 tool calls is selected. For more complex tasks (multi-step research, combined data queries) a higher value makes sense. Higher values cost more time and credits, so adjust them to your use case.

Assistant Check

The Assistant Check in the right sidebar continuously verifies that the basic building blocks of your assistant are in place:

• System prompt defined — the instructions are not empty
• Description added — your description in the General section is filled in
• Tools enabled — at least one tool is selected
• Knowledge sources linked — either uploaded files or a connected data pool