Review the record.
Build the work.

A case can stretch across documents, email, spreadsheets, interviews, and hours of recordings. The work is finding what matters, understanding how it fits together, and keeping the source close enough to check.

RecordBench brings that work into one place. Search the records, follow a transcript back to the recording, ask questions with citations in view, and turn useful findings into notes and draft work product.

It’s grounded in the work of federal criminal defense, with room to be useful to other legal and investigation teams facing the same problem.

Local control is central to the project. RecordBench is designed to run on infrastructure an office controls, with local models powering its bundled AI workflows.

AI can help find a passage, work through a question, or make a recording easier to navigate. The useful part is being able to inspect the support, see the gaps, and carry the work forward. Answers and transcripts are review aids; the person doing the work makes the call.

Intake now keeps a receipt of selected files and skipped items, with resumable uploads. Folder browsing and matching-file views help make sense of a delivery while preserving each source and its references.

Focused questions and broader investigations share a cited conversation. Investigations can follow names, dates, and phrases back into the record. Full-text review applies a question to a frozen collection of extracted text, keeping coverage gaps and human review visible; a completed run does not mean every page or relevant fact was found.

Recordings sit alongside timestamped transcripts and clips. People, places, events, and evidence connections retain support from the originals, including uncertain dates and competing accounts. Notes and editable Reports carry useful findings forward, and shared editing preserves unsaved work when changes conflict.

Recent work also makes limitations easier to inspect: email attachments need separate review, changed source collections can call for a fresh answer, and generated claims still need checking against the originals. An export check flags saved Reports needing attention before a complete matter-bundle download.

The alpha’s review workspaces remain temporary. Export and verify work you need to retain before closing a matter or its retention period expires.

The longer-term goal is to give a team a lasting place to develop its understanding of a case, from the first discovery delivery through the work it needs to take away.

The current people, events, and evidence-connection tools are early steps toward that goal. The next work includes more durable matter memory, deeper source-grounded investigation, and better ways to exchange work with the tools an office already uses.

Practical installation, recovery, and broader release validation remain part of that development work. The longer-term goals extend beyond what the current alpha provides.