Skip to main content
Home > “Admin Menu” > “Connectors”

​
Connectors

Connectors extend VARIOS AI with direct integrations to external systems, data sources, and services. They make it possible to connect assistants with enterprise applications, search services, databases, documentation platforms, or custom APIs. This allows assistants not only to access their language model and internal knowledge databases, but also to retrieve current information, perform actions in third-party systems, or process structured data from external sources. Connectors can be integrated into assistants selectively and controlled via group permissions.
Info: Connectors are provided as tools in assistants and can be used by the model during the conversation.
Note: Which connectors may be used is controlled via the respective configuration as well as via groups and assistants.

​
Functions at a Glance

  • Central Management: Connectors are created and managed centrally in the admin area.
  • Integration into Assistants: Connectors can be assigned specifically to individual assistants.
  • Group Control: Access to connectors can be restricted via groups.
  • Wide Range of Integration Options: Supported options include search services, databases, webhooks, Microsoft 365, OpenAPI, and MCP.
  • Cost Overview: Connector costs can be evaluated in the dashboard depending on the configuration.

​
Configure Connectors

When using connectors in VARIOS AI, the following steps are typically relevant:
1

Create Connector

Create a new connector in the admin area with a meaningful name and an image.
2

Create Description

Describe the tasks and functions of the connector.
3

Select Connector Type

Select the connector type:
  • Under Code, you will find our prebuilt solutions. Fill in the available options and arguments.
  • Under OpenAPI, you will find your created OpenAPI configurations
  • Under MCP, you will find your created MCP configurations
4

Set Data Protection (DLP)

Under Data protection (DLP), decide per direction whether the connector receives the values Same as chat or as Real data under Data to the connector, and whether Results from the connector reach the model pseudonymized or unchecked. See Data protection (DLP) per connector for details.Optionally enable Human-in-the-Loop: Execute only after user confirmation here. The user must then approve the connector call initiated by the AI model and can adjust the execution content manually. The option can only be enabled together with Real data.
5

Control Access via Groups

Define which groups may add the connector to their personal assistants.
6

Assign Connector to an Assistant

Open the editing view of the desired assistant and add the connector.

​
Data protection (DLP) per connector

While the DLP check is active in a chat, the language model only sees protected findings as pseudonyms. Two settings on the connector define how a call deals with that.
Human-in-the-Loop can only be enabled together with Real data. This way the user approves exactly what the connector actually receives.
Recommendations for Data to the connector / Results from the connector:
  • Search services inside your organization (Confluence, databases, Personio): Real data / Pseudonymized.
  • Microsoft 365 and Microsoft 365 Calendar: Real data / Unchecked. E-mails, calendar entries and recipients can only be found and used with their real values. See the note under Microsoft 365.
  • Public search services (web search, Stack Overflow, documentation services): Same as chat / Unchecked. Exceptions: for a SearXNG instance that searches internal sources and for Read the Docs with a private project, results stay Pseudonymized.
  • Sending e-mail (Microsoft Mail Send): Real data / Unchecked, plus Human-in-the-Loop.
  • Webhook: Real data / Pseudonymized, plus Human-in-the-Loop.
  • Image and video generators: Same as chat / Pseudonymized.
During the update, existing prebuilt connectors (type Code) whose settings are still at the default are adjusted as follows:
  • Real data goes to Microsoft 365, Microsoft 365 Calendar, Microsoft Mail Send, Personio, Zammad, Confluence, OpenSearch, Pixxio, MSSQL, MySQL and PostgreSQL, Databricks and Webhook.
  • Results reach the model Unchecked for Microsoft 365, Microsoft 365 Calendar, Microsoft Mail Send, Microsoft Bing Websearch, SearXNG Search, Web Crawl, Stack Overflow, Mintlify and Read the Docs. If a SearXNG instance searches internal sources or a Read the Docs connector accesses a private project, set its results back to Pseudonymized after the update.
Existing connectors with Human-in-the-Loop receive real data during the update. New connectors start with Same as chat and Pseudonymized.

​
References

The References tab in the editing view of a connector shows where the connector is used: Clicking an entry opens the assistant or the group. Personal assistants are not listed individually; their number appears below the list, for example “3 personal assistants”. For saved MCP connectors, the tab also lists under Signed-in users the users who have signed in to the MCP server. Use Disconnect to remove a sign-in.

​
Typical Connector Types

VARIOS AI supports different types of connectors for different use cases:

​
Search and Documentation Services

  • Confluence
  • Mintlify
  • Read the Docs
  • Stack Overflow
  • SearXNG Search
  • Web Crawl
These connectors are suitable for research in documentation, wikis, websites, or technical knowledge sources.

​
Enterprise Applications

  • Microsoft 365
  • Microsoft 365 Calendar
  • Microsoft Mail Send
  • Personio
  • Zammad
These connectors enable access to emails, calendars, files, tickets, or HR data.

​
Databases and Search Indexes

  • MSSQL
  • MySQL and PostgreSQL
  • OpenSearch
  • Databricks
These connectors are suitable for structured data queries, reports, and search applications.

​
API and Automation Integrations

  • Webhook
  • OpenAPI
  • MCP (Model Context Protocol)
These connectors are especially flexible and are suitable for custom integrations, process automation, and connecting proprietary services.

​
Media and Generation Services

  • OpenAI Image
  • Google Gemini Image
  • Flux
  • Replicate.com
  • OpenJourney
  • Runway
  • Google Veo Video
  • xAI Imagine Video
These connectors enable image and video generation directly from VARIOS AI.

​
Typical Use Cases

  • Research in enterprise knowledge via Confluence, Microsoft 365, or documentation platforms
  • Retrieval of structured data from SQL databases or OpenSearch
  • Integration of internal APIs via OpenAPI or webhook
  • Connection of external tools and services via MCP
  • Automated ticket and support processes with Zammad
  • Email workflows with Microsoft 365
  • Image and video generation for marketing, training, or content production