Skip to main content

AI

Overview​

This section shows how to use the ADITO AI platform in practical implementation scenarios. It is the entry point for working with its APIs, understanding the currently supported capabilities, and applying the platform in CRM-related automation and assistant use cases.

The ADITO AI platform provides access to a self-hosted large language model running on ADITO Cloud infrastructure in Germany. No data is sent to external AI services.

ADITO AI, ADITO Intelligence, and ADITO AI Runtime​

Three related terms appear throughout this section and are easy to mix up. This is how they relate:

TermWhat it means
ADITO AIThe umbrella term for everything AI in the ADITO platform. It covers all AI products, features, and infrastructure, including the two below.
ADITO IntelligenceThe product layer of ADITO AI: the chat interface and the AI features that users interact with directly.
ADITO AI RuntimeThe backend of ADITO AI: the self-hosted infrastructure that runs the AI models and serves them through OpenAI- and Anthropic-compatible APIs. See AI Models.

ADITO Intelligence and the ADITO AI Runtime are both subsets of ADITO AI: ADITO Intelligence is what users see, and the ADITO AI Runtime is what runs behind it.

Background and updates

For product background and release context, see the blog posts AI is Part of ADITO Cloud and ADITO-LLM Now Supports Thinking.

What's in this section​

  • AI Models: the model kinds served by the ADITO AI Runtime — text generation, code generation, embeddings, and speech-to-text. Text Generation also covers the ADITO-LLM API reference and what you can build with it.
  • Compliance: security, data handling, transparency, and ethics.
  • Prompting Guide: how to write good prompts.
  • Token Usage Estimation: a practical method for estimating how many tokens a use case will consume.
  • Building an Agent: how to build a basic agent loop in JDito today with function calling.
  • Building a RAG Use Case: how to combine vector search over CRM data with the ADITO-LLM so answers are grounded in your own records.

Getting started​

Managed systems

Managed systems are preconfigured out of the box since core version 2025.1.0 and require no additional setup (see Configuring the AI alias).

For chat, embeddings, and speech-to-text, the API is OpenAI-compatible: point any OpenAI client library at https://ai.adito.cloud and supply your API key. Every managed system receives a system API key at startup. If you need a personal key, contact the AI team.

note

For endpoint details and usage examples, see Text Generation. To use a coding agent such as OpenCode, Claude Code, or Codex instead, see Code Generation; it uses a different interface.

FAQ​

What is an agent?

Unlike conventional LLM applications such as single-turn chatbots or classification models, agents are autonomous systems that execute multi-step tasks with minimal human oversight. They combine reasoning, memory, tool use, and workflow management to complete complex goals independently. See Building an Agent for how to build a basic agent loop today.

What is RAG?

RAG (retrieval-augmented generation) grounds a model's answer in your own data. Instead of relying on what the model learned during training, the relevant records are retrieved first — by meaning, using vector search — and passed to the model as context. See Building a RAG Use Case for the full pipeline.

What is an MCP server?

An MCP server (Model Context Protocol) connects tools and data sources to the AI in a structured way. This allows web services, databases, or third-party systems to be called by the model when needed.

Where does my data go?

Your data stays inside your separate environment and database. All communication is routed exclusively to ADITO self-hosted AI models in Germany. There is no connection to public or external AI services such as OpenAI, Azure OpenAI, or Alibaba Cloud. See AI Compliance for the full data handling and retention policy.

Which AI does the platform use by default?

All AI features in the ADITO platform — including the CKEditor (HTML editor) AI Assistant — communicate with the ADITO AI Runtime by default, which is self-hosted on ADITO infrastructure in Germany. No data is sent to external or public AI providers unless a different AI alias is explicitly configured.

Can ADITO-LLM read scanned documents or images (OCR)?

Yes. The chat endpoint takes images alongside text, so a scanned invoice or a photographed delivery note can be transcribed with a normal chat request. There is no separate OCR service and no extra API key. Combined with guided_json, one request returns the extracted fields ready to be written to a record. See Image input and OCR.

Is the API compatible with OpenAI client libraries?

For chat, embeddings, and speech-to-text: yes. Change only the baseURL to https://ai.adito.cloud and supply your ADITO apiKey. All other parameters and methods work the same way. Code generation also supports OpenAI-compatible clients (e.g. OpenCode) alongside the Anthropic Messages and OpenAI Responses APIs used by Claude Code and Codex; see Code Generation for per-tool setup.

How do I get an API key?

Managed ADITO Cloud systems receive a system key automatically at startup.

I cannot find the API key in my cloud system — what should I check?

Two things to check:

  1. ADITO Designer version 2025.1.0 or later is required to view and edit the AI alias configuration under System Configuration.

  2. The Operator creates the AI alias configuration on first startup only. If a record already exists, it is left unchanged — endpointUrl, token, and requestParameters included. Manual edits to those fields survive restarts, but a broken or outdated entry will not be fixed automatically.

To reset to the Operator defaults, delete the AI alias in ADITO Designer, then restart the system (OFF, then ON). The Operator recreates the AI alias configuration on the next startup.