Skip to main content

Conclusion Models (Sandbox)

In the Sandbox environment, embraceableAI currently provides the e2-light Conclusion Model to explore Cognitive Control and structured reasoning processes. All Sandbox models use Cognitive Control for structured conclusions with full transparency and policy compliance.

Additional and more powerful models (including non-Sandbox/production variants) are available after consultation with embraceableAI and can be added to this overview in the future.

Sandbox Model Overview

The Sandbox currently exposes one model for testing and getting to know the platform:

ModelEnvironmentBase Language ModelBest For
e2-light / e2-light-patSandboxManaged by embraceableAIGetting started, testing and evaluation of the Conclusion API and UI

Note: The model name may differ for technical reasons (e.g., e2-light vs. e2-light-pat), but the underlying model is the same. To retrieve the specific model name that can be used with your token, query the models endpoint endpoint. For PATs, the model name will end with -pat.

e2-light Model Process

The e2-light model follows a structured reasoning process that ensures compliance, traceability and high‑quality conclusions.

General Process Flow

e2-light executes conclusions in the following phases (the corresponding output type in the Conclusion API is written in parenthesis):

  1. Compliance Check (malicious_intent) - Initial validation of inputs and policies
  2. Planning (planning) - Understanding the task and breaking it down
  3. Constraint Extraction (constraint_extraction) - Parsing and understanding policies and constraints
  4. Step-by-Step Task Resolution - Executing the solution through discrete steps
  5. Finalization and Reflection (finalize) - Producing the final answer with reflection on the process

Execution Step Monitoring

During the Step-by-Step Task Resolution phase, each execution step is monitored and validated through a structured cycle:

  1. Approach (approach_generation) - Determining how to solve this specific step
  2. Execution (solve) - Performing the step
  3. Reflection (reflection) - Checking whether constraints and policies were adhered to
  4. Result (result) - Producing the final answer for this step

This monitoring ensures that every step maintains compliance with the defined policies and constraints before proceeding to the next step.

Beyond the Sandbox

For productive use cases, stricter data residency requirements or higher performance and quality, embraceableAI provides additional model variants and deployment options (including non-Sandbox/production environments). These can be enabled for your account after consultation and will be documented in extended model overviews in the future.

Getting Started

1. Use the Sandbox model

Start with the e2-light Sandbox model to familiarize yourself with the Conclusion API and UI.

2. Set Up Authentication

  • Get API tokens for the Sandbox
  • Configure the e2-light model in your requests

3. Additional models (on request)

If you need additional models or production deployments, contact embraceableAI. Individual and production-grade models can be provisioned for your account and will then appear in the responses of the models endpoint.

Support

For custom requirements or specific problems, the embraceableAI Solution Team is happy to assist. The entire stack can also be individually provisioned, or custom models and capabilities can be used to meet your specific needs.