willchen96 teaches Mike to read the case law
Mike stops being a document assistant and starts doing genuine legal research - grounded in real precedent, not invented citations.
This release is a big one, and the marquee addition is case-law research wired directly into the assistant. It connects to CourtListener - a large free public database of US court opinions - and, crucially, follows a disciplined research order rather than free-associating. The same release ships a cluster of other upgrades:
- Case-law research: the assistant verifies a citation is real before trusting it, pulls the cases behind it, keyword-searches inside opinions, and reads full text only when the question actually demands it - designed to keep answers anchored to precedent that exists.
- Document version history: files are now tracked as versions instead of flat documents, so you can upload a new version, drag-and-drop one in, review the version list, and delete with a confirmation step.
- Interruptible answers: you can now cancel a response while it is still being generated.
- Security hardening: each user's stored provider keys are encrypted, third-party opinion text is scrubbed before display, and backend data tables are locked down.
- A broad visual refresh ("liquid glass") applied across menus, panels, and settings.
So what If you have ever been burned by an AI citing a case that doesn't exist, this is the fork feature worth watching.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?