10 February 2025
Layout-aware drafting
Improvement We continue to tweak our new layout engine (launched in January) that allows users to insert clauses and redraft text while respecting the layout of the opened document.
We are proud to say that the majority of legal documents "in the wild" are now handled with resonable layout settings. Due to the nature of legal documents — most tend to have mediocre to very bad layouts, technically speaking — it will never be possible to provide perfection, but ClauseBase goes significantly beyond what everyone else (including Microsoft's CoPilot) is doing with plaintext-only.
Read more about this in our news flash.
AutoCheck
New major feature In the Write & Rewrite module, users can now initiate an AutoCheck or AutoPolish action, for zero-effort rewriting of existing documents.
In AutoCheck, users must simply select which parties they are acting for (e.g., the customer vs. the supplier, or the employee or the employer).
With the help of the built-in legal intelligence of the LLMs, ClauseBuddy will then present a fully automated markup of the selected text. For example, the following suggestions are made if acting for the Licensee:
This works best for documents that are well-known to LLMs and are not too jurisdiction-dependent — for example, the various types of commercial agreements. If any guidance must be provided to the LLM, or background knowledge is required for the LLM to understand the contract, then playbook-based or checklist-based workflows must instead be considered.
AutoPolish
New major feature ClauseBuddy now allows users to perform automated proofreading of texts, with automatic changes being applied.
With Only correct obvious errors, ClauseBuddy will instruct the LLM to only correct, well, the very obvious errors — similar to what a senior would ask from a junior, as a last review before a document is sent to the counterparty. For example, the following changes are then made:
The use case of this feature is to go beyond Word's standard spellcheck and grammar check, which does not truly "understand" text and only focuses on well-known problems.
The Correct errors and optimise wording goes beyond this. With this option, the LLM will also propose various optimisations. This is the equivalent of asking a native speaker or editor to go through your final text — expect many things where you are positively surprised, but also expect many changes that you don't agree to.
Checklists
New major feature Any reviewing rules that are created can now also be used for "checklists".
Checklists are intended for legal professionals and/or legal documents for which the fully automated reviewing process is not desired. Those fully automated reviews are most suitable for reviewing relatively short, easy-to-medium complexity documents that can be sufficiently standardised so that rules ("playbooks") can be written for a machine to understand.
Conversely, complex deals and highly variable documents are not yet the primary candidates for fully automated reviews. In addition, many legal professionals don't trust those reviews yet, and instead want to go through documents themselves.
For those situations, checklists are a perfect balance between the traditional reviewing process (no LLM involved) and the new-style reviewing process (everything through LLMs). The idea is that legal teams prepare a standard checklist in advance, and that legal professionals then go through their document and the checklist, checking and annotating the checklist as you go.
Individual items of a checklist can easily serve as the basis for a rewrite with the LLM.
Checklists can be saved and reloaded at any point.
The checklists can be completely customised.
The output of automated legal reviews can serve as a new (partially completed) checklist.
We provide 32 sample checklists of various legal documents, to give legal teams a starting point.
Sample checklists & playbooks
New feature We provide a list of 32 sample playbooks that can serve as starting point for reviewing sets and/or checklists.
The idea is that those checklists are discussed internally within the legal team, customised where necessary (mostly by deleting) and then uploaded to ClauseBuddy, where they will be immediately converted to a "review category", ready to be used for automated reviewing and/or checklist-based reviewing.
Automatic import of existing playbooks
New feature Existing playbooks can now be converted into ClauseBuddy's playbook-format, ready to be used in automated reviews or checklist-based reviews.
Either users can upload an entire DOCX file, if the document has a structure that resembles ClauseBuddy's own samples (i.e. with a group title in Heading 1 style and the actual rule in Heading 2).
Or users can select text from their currently opened Word-file and have the LLM extract the rules intelligently.
Custom report layouts
New major feature It is now possible to use custom layouts for the reports exported from an automated review, as well as various other exported documents (such as summaries and checklist reports).
As an admin, you can actually upload multiple templates (e.g., for different departments or clients) through the Settings > Export Templates.
All those documents you upload there, will become part of the dropdown list for end-users.
New types of review rules
New major feature There are two new types of elements that can be added to a review category:
Information extraction allows you to ask the LLM to simply extract data (e.g., the name of the signatories, or the geographical location of the real estate that’s being sold, or ...). The LLM will report this data as the first elements in its output (also exported to your report), and even tell you if there are ambiguities in what it found.
Literal text matching allows you to look for literal matches of text, e.g. literal fragments of some legislation that must (or must not be) present in the document. Background: these rules are NOT passed to the LLM — instead, they are executed through traditional text matching — because LLMs are very bad in searching for literal text in a large document.
Customisation of predefined prompts
New major feature Large organisations will be happy to hear that they can now customise the predefined redrafting prompts.
As an admin, you can now configure which redraft examples should be shown to the end-user at the bottom of the “Redraft” panel in the Write & Rewrite panel.
There are quite some configuration options: you can for example show/hide the ClauseBase system defaults, or show different examples for different departments, and/or have firm-wide examples. You can also configure the ordering of each of those.
Personal truffle baskets
New feature Organisations can now add personal truffle baskets for everyone (also future users).
Improved quality during reviewing
Improvement After many test sessions and with the help of the feedback from many customers, we have been able to increase the quality of the automated reviews being done by the LLMs. In particular:
You can now have reviewing sets of unlimited length. Previously long lists often caused the LLMs to get overloaded so that their responses were poor, but this has been resolved now (at the expense of more "token" usage and somewhat slower responses).
You can set a "quality" indicator (through the ... icon in the top-right corner during review). This can be useful if individual requirements are very complex to understand for the LLM. Setting the quality to "high" will cause ClauseBuddy to be more cautious about what it sends to the LLM.
We have rearchitected the way clauses are being referenced. Previously, the LLM would often refer to too many paragraphs when it got confused due to the amount of information to process.
Last updated