Eclipse ADL: Standardized Language for Design and Control of AI Agents

The Eclipse Foundation has introduced an open, dedicated description language for planning, defining behavior patterns, and controlling AI agent systems. Agent Definition Language (ADL) describes agent systems in simple, standardized categories and terminology, which are intended to enable structured deployment and regulated operation.

According to Eclipse's announcement, such an open and transparent standard is not yet available on the market, allowing companies to introduce and maintain vendor- and model-independent agent systems.

Linguistically Guided Use Cases

In its concept, ADL separates the definition of agent systems from concrete prompting, with the design happening in the business departments using ADL, while ideally a compiler automates the implementation into prompts and integrates the tools needed to solve the task.

In terms of terminology, the language categorizes agents into use cases, within which various scenarios occur. Here, users interact with agents, or agents interact with each other. The formal definitions include:

• UseCase Name: A concise, descriptive identifier that uniquely identifies the use case.
• Description: A detailed explanation of the situation or user requests.
• Steps (optional): A sequence of steps the agent must perform before providing the final solution.
• Solution: The recommended solution to fulfill the request.
• Alternative Solution (optional): An alternative solution the agent should try if the first suggestion fails.
• Fallback Solution (optional): This final fallback option is intended to prevent the agent from getting stuck in a loop, repeatedly offering the same suggestions.
• Examples (optional): This section provides additional context.

In code:

### UseCase: password_reset
#### Description
Customer has forgotten their password and needs to reset it.

#### Steps
- Ask the customer for their registered email address.
- Send a password reset link to the provided email address.

#### Solution
Guide the customer through the password reset process defined on the webpage https://www.example.com/reset-password.

#### Fallback Solution
If the customer cannot access their email, escalate the issue to a higher tier of support.

#### Examples
 - I forgot my password.

Other elements include flow options, with which users can implement decision trees:

[option 1] command
[option 2] command 2

To try it out, Eclipse offers a playground, which already includes a first use case, “Automobile Example.” More in-depth information can be found in the technical documentation.

Part of an Open Agent Platform

ADL integrates as a concept into the Eclipse agent platform Language Model Operating System (LMOS), which includes two further components: the ARC Agent Framework and the LMOS platform. ARC offers a JVM-native framework with a Kotlin runtime environment for developing, testing, and extending AI agents with a visual interface. This already implements ADL.

The LMOS platform is still in the alpha stage and is intended to serve as an open orchestration layer for the infrastructure, based on open-source tools from the CNCF (Cloud Native Computing Foundation) ecosystem. LMOS is already in use at Telekom as the basis for the chatbot “Frag Magenta.”

CLC 2025: Conference for DevEx and Platform Engineering

The Program of the CLC, from November 18th to 20th, 2025 in Mannheim, covers all topics related to Platform Engineering and Developer Experience. Tickets and further information on the CLC website.

(who)