Every step from
upload to red flag.

Upload filings, contracts, CIMs and data exports as PDF, DOCX, TXT or CSV.

Upload up to 25 files at once, screen them together, and follow each file’s status as it goes.

Use the web portal, call the REST API, or let an agent in Claude or ChatGPT screen documents over MCP.

When a deal room has no industry set, each document is classified and scored against the matching ruleset: M&A, banking, pharma or aviation. The report says when a ruleset was picked this way, so you can confirm it.

Financial, legal, HR, sanctions and AML, merger control, IP, AI governance, data privacy and more. See each category.

Seven categories can block a deal. If any one of them scores 70 or above, the deal is flagged for mandatory review, however clean the rest looks.

One close read of the whole document, for contracts and filings up to 130,000 characters of text.

A 300-page filing isn’t cut to fit. It’s split into chunks and every chunk is scored. 25 a month on Growth, unmetered on Pro and Enterprise.

Social security numbers, card numbers, emails and phone numbers are redacted before text reaches the AI model. Company and person names stay, because sanctions screening needs them.

Quotes and figures in each finding are matched back to the document text. A finding whose quote can’t be found is marked unverified, not quietly kept.

Findings become a list of what to verify, most urgent first and grouped by diligence workstream, so each specialist sees their share.

Mark findings reviewed, flagged or dismissed as your team works through them.

Ask a question about a scored document, or get a summary, answered from its own text. Available for 30 days after the document is scored.

Compare two or more documents to catch figures, dates and parties that disagree, such as a CIM against the audited accounts. Comparisons are saved until you delete them.

Download a formatted report as PDF or DOCX to share or file.

Machine-readable output for your own tools, models and pipelines.

Download every report from a batch in a single file.

Every download is stamped with who downloaded it, their role and the time in their own timezone, so a copy filed in a data room identifies itself.

See a room’s risk by category and workstream across every scored document, and how many documents have been scored. There is deliberately no single deal score, because an average hides the one category that matters.

Invite people as analysts or viewers. Viewers can read results but can’t upload.

Give the same person a different role per room: analyst in one, viewer in another, clean team in a third.

Team members accept a room’s confidentiality terms before they can open its documents or its risk summary.

The account owner can review each room’s own audit log of who did what, and when.

Ask Kai how to use the platform or how to read a report. Kai has no access to your documents.

For competitively sensitive material, kept in a separate database schema with its own key and room roles. The deal team sees nothing until a lead or counsel releases it, and every change, release decision and download goes into a hash-chained audit log. It’s software, not clean-team services.

Upload, screen, check status and download reports from your own systems, server to server.

Four screening tools for Claude, ChatGPT and other MCP clients: screen a document, fetch an analysis, submit a batch and check its status.

Pull scored results into your CRM or BI tools.

Risk by category across every room on the account, through the API.

Extracted text and findings are encrypted at rest under a key derived for your account alone.

The app limits every query to your account, and Postgres row-level security enforces it a second time.

Documents are processed by Anthropic’s Claude over encrypted TLS and are never used to train AI models.

Every upload, analysis, report view, download and change is recorded: who, what and when.

A scored document’s extracted text is deleted 30–31 days after analysis. Your reports stay.

Export your account’s data, or request its deletion, through the API.