If you run a solo or small firm, your real product is the body of context you've built around each matter. The pleadings, the discovery, the deposition transcripts, the client's email about the third inconsistent date in their timeline, the opposing counsel's last letter that hinted at a settlement window. That context is the reason the case moves at all, and it's the part of the practice that lives most in your head and in folders nobody else can navigate.
AI notes earn their keep when they hold that context for you and let you read across it on demand. Not "ask AI a legal question." Ask the agent: pull every reference in this matter to the date the contract was signed, summarize what the client said in the intake compared to what the deposition shows, draft a motion outline using the structure of the one I won six months ago. That's the version of AI that fits a practicing lawyer's day.
A note on scope before the rest: nothing in this post is legal advice on tool selection, ethics, or unauthorized practice. Talk to your bar regulator about what you're allowed to put in a cloud vault and what disclosures you owe clients. The mechanical capability is here. The compliance posture is yours. Other regulated-document workflows worth a look: AI notes for accountants and CPAs, and on the personal side, notes for legal documentation for individuals tracking their own paper trail.
A matter file that's actually a file
The shape that holds up across matter type is roughly: one top-level page per matter, with sub-pages for the engagement letter, the pleadings, the discovery log, the deposition prep and transcripts, the correspondence log, and the running motion drafts. Capy supports unlimited page nesting, so a heavy matter can fan out into per-issue or per-witness pages without you committing to a depth in advance.
Everything lives in your vault as plain markdown notes. That matters because the agent can read across all of it in one query. When you're prepping for a hearing and the judge is likely to ask about a specific contention, you don't have to open six tabs. You ask the agent: pull every reference across the pleadings, depositions, and exhibit log to the alleged breach date, and tell me where my strongest support is and where the record is thin. The agent reads the file and writes the answer.
Discovery that becomes searchable instead of opaque
A typical matter drowns in PDFs. Bates-stamped productions, deposition transcripts, expert reports, exhibits, opposing counsel correspondence. Most of those PDFs sit in a folder labeled "discovery" and get skimmed once.
Drop the PDFs on Capy pages. They auto-convert to markdown via docstrange, which means the agent can read every page — not just keywords — and treat the production as searchable text the same as any other note. Ask the agent to find every mention of a specific witness across a thousand-page production. Or to surface every exhibit that touches on the calendar window between two dates. Or to pull every Q-and-A in a deposition transcript that goes to a particular issue.
The conversion runs once at upload. From then on, the document stays searchable. This is what makes "chat with discovery" work as a real workflow rather than a parlor trick — the agent isn't running OCR on every query; it's reading text it already has.
The transcripts and productions stay where you put them. The vault is a working layer on top of the underlying documents, not a replacement for the system of record your firm uses.
Recording client meetings with speaker labels
Client intakes and witness prep sessions produce information that gets lost the moment the call ends. The client mentions a fact in the intake that turns out to be load-bearing six months later. The witness slips in a date during prep that contradicts what they said on the record. Your contemporaneous notes capture some of it. The recording captures all of it.
Record the meeting inside Capy. The transcript comes back with speaker diarization — labels like "Speaker 1: …" so you can tell who said what. Park the recording on the matter's intake or prep page. Ask the agent to draft a summary, pull every concrete factual claim, and flag any answer that contradicts something already in the file.
The transcript also gives you something for the file that's harder to second-guess later. If the client tells you on a call that they did read the contract before signing, that's now in the record with their own words and the date.
The usual caveats apply: get the client's consent to record per your jurisdiction's rules, store the recording per your professional standards, and don't pretend a transcript is a substitute for a deposition. The mechanical capability is the upgrade.
Drafting motions from your past motions
Motion drafting is the kind of work that compounds badly. You won a motion to compel six months ago and you're writing a similar one this week. The structure that worked is in a folder you've forgotten the name of, and the temptation is to start from a template instead of from your own past win.
Tell the agent: read the four motions to compel I've filed in the last two years, identify the one that drew the cleanest ruling, draft a v1 of this matter's motion using its structure, and pull the language about specificity from the order I cited successfully last time. You get a draft that descends from your own work. You spend the saved time on the parts that genuinely are new — the specific facts and the legal hooks unique to this matter. The agent acts on the document directly — the same idea covered in Claude Code for documents, but for matter files instead of code.
If a relevant case has come down since you last briefed the issue, the agent's web_search tool can pull the published opinion into the same page with a source URL so you're not toggling between a browser and your notes. You're still reading and citing the case yourself; the agent is the research assistant, not the lawyer.
Court prep that doesn't restart from zero
The night before a hearing is the wrong time to discover that the most useful piece of testimony is buried at page 312 of a deposition. Court prep that holds up is mostly about being able to read across the file fast enough to catch the contradictions and surface the strongest support.
A working setup: have an inline database called "hearing prep" inside the matter's page, with rows for issue, your strongest support, the opposing position, and where in the file the relevant passages live. The database lives directly in the page via the :::database::: directive, so you don't switch to a separate prep tracker. Build the rows during the week, not the night before. Ask the agent to read across the depositions and pleadings and propose entries; you confirm or edit.
When you walk into court, you have a structured read of where the record actually is — not a vibes-based recollection.
A correspondence log that's actually useful
Most lawyers keep a "correspondence log" that's really an email folder. The folder is fine for retrieval; it's bad for pattern recognition. When opposing counsel softens their tone in week six and then hardens again in week eight, the shift is meaningful, and you'll usually only catch it in retrospect.
A working setup: forward every meaningful exchange into a Capy page tagged with the matter. Periodically ask the agent to read the log and summarize the trajectory — where opposing counsel started, what shifted, where they're likely positioning. The summary won't replace your read, but it'll catch the shifts that are easy to miss in real time.
What this isn't
Capy isn't a practice-management system. It doesn't track time, generate invoices, or manage your trust account. The structured-data side of the practice still belongs in the tools that handle structured data. Capy is for the unstructured side — the matter file, the correspondence, the prep, the drafts — which is the part of the day that's actually slow.
It's also single-user by design and cloud-hosted. One lawyer, one vault, on our servers. If your jurisdiction or matter type requires keeping data on your own hardware, or a multi-user firm workspace with role-based access, this isn't the right fit and we'd rather you know up front. For most solo and small-firm lawyers, the cloud vault with one integrated agent is a strict upgrade over the current reality of files-in-Drive plus a billing app plus a notebook.
A small first test
Take one active matter. Load the engagement letter, the most recent pleading, one deposition transcript, and any client correspondence on a single page in Capy. Ask the agent to write a one-page status summary of where the matter stands and what the highest-leverage next step is. If the summary catches a tension between two of those documents that you hadn't surfaced explicitly yet, you've got a sense of what the agent does across the rest of the file. The same drafting-from-past-work pattern shows up in our writeup of contract negotiation with AI notes.
Try Docapybara free. Load one matter and see whether the summary surprises you.