Definely AI FAQs

In May 2025, Definely launched the Word-native Legal AI assistant, Enhance. Check out below for all your questions about Definely's approach to AI as well as Enhance itself. If we haven't answered your question, please do get in touch.

find your answers

AI FAQs

Is any data stored?

Definely has a data retention period of 14-days after which the data is cleared. During this period, we store the data but do not touch it. If a customer highlights a particular problem before the end of the 14 days then, with permission, we will access the data to investigate the specific issue.

Where does my data reside?

Your data will reside in Definely’s private cloud tenancy in AWS. Currently, deployments to AWS UK and EU-Central regions are available.

Who has access to our data? Can you view our data?

Only authorised support engineers will ever view data and only with your explicit permission for debugging purposes.

‍

Will my data be used to train an underlying AI model?

No. The core language models used by our system are accessed through a deployment on our private Azure instance.

We have a data retention policy of 14-days after which the data is removed.

‍

Can clients use their own hosted LLM as part of an Enhance or Cascade implementation?

No - Enhance is a cloud product managed by Definely. This simplifies implementation, enables further development of AI features and reduces the maintenance costs for our customers.

How is the quality and accuracy of Definely's output evaluated?

Model performance is evaluated and improved using Definely’s proprietary legal dataset and in-house expertise. This ensures LLM outputs are unbiased and quality of the product is in line with customer expectations.

‍

More about Definely AI

find your answers

Enhance FAQs

Is there a limit on the number of documents you can link to answer a question?

There isn’t a limit to the number of documents that can be linked but performance may be dependent on the length of the documents. For example, five small documents could be smaller than one large contract. More thinking time is required the more documents you add.

We’ve linked documents using Definely Draft. Will we be able to query across those linked documents?

Lawyers will have to link documents again in the Virtual Assistant if they have been linked in other Definely products. We are currently reviewing how to make this more cohesive across our product suite.

‍

How do I provide feedback on responses from the Virtual Assistant?

Lawyers will be able to like or dislike individual responses from the Virtual Assistant. To help us gauge users willingness to use the feedback function, Definely will track clicks on likes or dislikes. Definely will not track ID’s nor content of the responses getting liked or disliked.

Can I ask follow up questions for deeper document analysis?

Yes - lawyers are able to have a conversation with the Virtual Assistant, drill into responses and ask follow up questions. This allows for deeper analysis of documents and further refinement of answers.

‍

If I close the document and open it again later, will I be able to access previous chats?

No - Definely doesn’t store any chats. The history is cleared once the document is closed

‍

How does “Summarise” prioritise clauses?

The underlying model chooses what it deems to be the most important clauses in a contract. Lawyers will be able to filter to select which clause to summarise.

‍

How does Definely’s agentic framework work?

How does Definely’s agentic framework work?

Enhance employs multiple models to form an agentic framework. Each model - or agent - is specialised to cater for specific tasks such as summarisation or to support drafting. Definely Enhance analyses the task, creates a plan and assigns agents to handle different parts of the task. Agents collaborate and gather information passing it along until the task is completed.

‍

The performance of agents is evaluated and improved using Definely’s proprietary legal dataset and in-house expertise.

‍

What is the underlying technology?

Definely uses knowledge graphs, relationship extraction and large language models (LLMs). Our proprietary technology breaks the contract into clauses and definitions. This information is used to build a knowledge graph, a graphical representation of entities mentioned in the text, mapping their relationships.

We then use a network of models to build an agentic framework within our private network to augment the knowledge graph.

More about Enhance

find your answers

Cascade FAQs

What kind of impact can Cascade identify?

Cascade identifies both cross-references and semantic impacts throughout your contract. When you modify a definition or clause, it flags all directly related sections—whether through explicit cross-references or semantic connections—as first-order impacts. It then traces how these changes affect other dependent clauses, identifying second- and third-order impacts. This comprehensive approach ensures you see not only the immediate effects of a change but also how it ripples through your entire contract.

How does Cascade detect knock-on effects in a contract?

Cascade automatically analyses the structure and language of your document in Microsoft Word. When you make a change - such as editing a definition or clause - it scans the entire contract to identify where that change creates first-, second-, or third-order impacts elsewhere in the document. These are then flagged in the Cascade sidebar for review.

What is the underlying technology?

Cascade is built on Definely’s Agentic Framework and leverages our proprietary technology that breaks the contract into clauses and definitions. It also uses embedding technology and Large Language Models to rapidly detect the semantic and structural implications of edits.

More about Cascade

See Enhance in action