From first prompt to the report a carrier asks for.
Six steps, designed to sit inside how a firm already works: no browser babysitting, no intercepting anyone's traffic, no new habit to enforce.
Tap any step to see it.
Confidentiality pre-check See it
Before text goes to an outside AI tool, Candor scans it for client identifying details: names, SSNs, account and policy numbers, dates, addresses, medical terms. It returns a redacted version to paste instead. Screening happens in the browser; the text itself is never stored. Try it below.
The scan runs on your device and reads names in their legal role context, not just against a word list, so it flags the client without redacting the caselaw. You paste the safe version; the original never leaves the machine.
AI use log See it
Each use is recorded against the matter: which tool, what task, who ran it, when. The firm keeps an approved tools list, and anything logged outside it is flagged: quiet visibility into what's actually being used.
A tool that is not on the approved list is flagged the moment it is logged, so a partner can see at a glance whether staff are using sanctioned tools, without watching over anyone's shoulder.
Verification See it
For research and drafting, Candor pulls out case citations and surfaces hedging language worth a second look, then checks each citation's existence against a public legal database. It's a prompt for the reviewer, not a verdict. Candor never claims a citation is correct.
Only the bare citation is sent out for the existence check, never the client's text. The hedging words are the AI's own signals of uncertainty, surfaced so the reviewer knows where to look hardest.
Attorney review & sign off See it
A supervising attorney marks the work reviewed and attests they independently checked the citations and assertions. Entries are append only: a change is a new record referencing the last, so the history stays intact.
The attestation wording is fixed and deliberate. Because entries are append only, a later correction shows up as its own record, so the trail reads like a ledger, not a document someone could quietly rewrite.
Reports See it
Candor generates a firm wide summary, and a renewal questionnaire mapped to the questions carriers ask, both built from the record and exportable to PDF. This is the artifact you hand across the table.
Same data in, same PDF out, so two people generating the report for the same period get the identical artifact. The renewal version is mapped in advance to the AI questions carriers now ask.
Policy management See it
Adopt a written AI use policy; staff acknowledge it with a timestamp and acknowledge again whenever it changes. The acknowledgements live in the same record as everything else.
When the policy changes, everyone acknowledges again, and the earlier acknowledgements stay on file. You can show not just the current policy but the history of who agreed to what, and when.
It files against the matters you already have.
Connect the system your firm already runs on, and AI use is recorded against real matters instead of names somebody retypes. Candor supports Clio, MyCase, Filevine, PracticePanther, and Smokeball.
Your administrator authorises it
Sign in on the platform's own site, the standard way. Candor never sees a password, and your firm can cut off access from your side at any time without asking us.
Matters are read, never kept
Candor reads the matter list when it needs it and lets it go. Your client list is not copied into our storage, and every time it is read the access log records it.
Facts without the client name
Pull the matter number, practice area, and status straight into a prompt so nobody retypes them. The matter title stays out by default, because titles usually carry the client's name.
One thing to check on your side: private integrations need a mid tier plan or above at most platforms, and the person connecting has to be an administrator.
Beyond the core loop.
Readiness score
A live measure of how completely your firm can document its AI use, with the gaps named, plus a per person view of who is using AI and whether it is supervised.
Court disclosure certificate
Generate a draft AI use certification for a filing, populated from the matter's record, for the attorney to review and sign.
Client consent tracking
Record that a client agreed to, or declined, AI use on their matter, and how that consent was captured. Where a client has declined, Candor says so before anyone logs an AI use against that matter.
Exception log
People work around the process. Screening gets skipped, an unapproved tool gets used late at night. Each one is recorded with a reason and raised for a partner, so the record shows the whole picture rather than the tidy part of it.
Supervision queue
What is waiting on a partner right now, how long it has been waiting, and where one person's AI volume is running ahead of the review behind it. Approve or send back from the queue.
Billing note
Mark whether an AI use affects what the client is billed, and note how the work was produced, at the time it happens rather than months later when nobody remembers.
Production packet
Two exports. The whole firm record for your own carrier or counsel, or one matter on its own for a court, opposing counsel, or a client, with every other client's activity left out rather than blacked over.
Tamper evident record
Every log is hash chained and encrypted at rest. A one click check proves nothing was altered, removed, or reordered.
Time on task, measured
Candor counts the time actually spent working with the AI and drafts a billing entry from it. What the same work would have taken a person is your estimate, never ours, and where you do not give one no comparison is recorded.
Take your record with you
Export everything a firm holds, sealed or readable. A sealed export restores exactly, hash chain intact. A readable one still carries the hashes, so it can be checked years from now with no Candor and no software of ours.
Candor can ride along where the work happens.
An optional browser companion puts the same checks inside the AI tools your team already uses: ChatGPT, Claude, Gemini and Copilot. Nobody has to remember to open a separate app.
Screen before you send
A quiet panel on the page redacts client identifying details before anyone pastes them into a chatbot. The text stays on the device.
Check the answer for real cases
When the AI replies, Candor reads the answer on screen and flags case citations that may not exist, checked against a public database of real court opinions. A prompt for the reviewer, never a verdict.
And point to the real one
Where a citation does not check out but a genuine case by that name exists, Candor shows you that case and its actual citation, with a link to the opinion. It closes the loop instead of leaving somebody to go hunting.
Log it in one click
Capture the tool, the task, and the matter without leaving the page. The entry files into the same sealed record as everything else.
What the citation check can and cannot tell you. A match confirms the case exists, not that it stands for what the AI said. No match may mean the citation was invented, or simply that it sits outside the public database. Either way, it points a reviewer to look. Every match carries a link straight to the opinion, and to your own research service if the firm uses one, because reading the case is the only thing that settles whether it says what the AI claimed. It never decides, and the attorney's own review is what carries weight.
It reads only what is on the screen, and only when asked. Nothing leaves the browser except the bare citations it needs to look up. No traffic interception, and no monitoring of your people.
Setting it up takes about a minute, and no IT department. Somebody at the firm signs in, opens settings, and presses one button. The browser hands the companion its credential directly, so there is no code to copy, no key to paste, and nothing to configure. Each machine shows up in the record by name, and any one of them can be switched off on its own the day a laptop walks out the door.
That credential can only add to the record. It cannot read the firm's history back, so a lost laptop cannot become a way into what the firm has logged.
The pre-check, running right here.
This is the real screener, running in your browser on this page. Type something with a name or a number in it; nothing you enter leaves your device.
What your team sees on the page.
A short walk through the check running inside an AI chat. This is an illustration of the browser companion, not a live tool.
Illustration. The real check runs on the page, in your browser.