Skip to content

TULIP Docs

Data Handling and Model Behavior

Data Handling and Model Behavior

Inference-Only Operation

TULIP-hosted models operate strictly in inference mode.

Model weights are static.
No fine-tuning or online learning occurs.
User input does not modify the model.
Prompts are not used to retrain or improve models.

Session Memory

Models process only the content provided within the current request.

There is no cross-session memory.
The system does not retain conversational state once a request completes.
Each API call is stateless unless the client explicitly sends prior context.

Data Persistence

TULIP does not store prompts or responses for training purposes.

The platform does not retain user input as part of model behavior.
Any persistence of prompts or responses depends on how the calling application handles logging or storage.
Users deploying applications on top of TULIP are responsible for reviewing their own data storage policies.

All inference is performed locally within TU Delft infrastructure.

No user data is transmitted to external model providers.
No data is used for model training outside TU Delft.