Redactly: AI Redaction for Lawyers

Solo and small-firm lawyers handle NDAs and contracts constantly, but without in-house redaction tooling the process is manual, slow, and risky. A single missed name or dollar figure is a real confidentiality breach they're personally liable for. Redactly covers the full document lifecycle. A built-in Draft NDA generator writes one from scratch, or you can upload an existing document. A one-click "Analyze this draft" button sends a generated document straight into the same pipeline a real uploaded document goes through, so there's no separate demo path, it's the same code either way. Both feed into the same five-stage pipeline: Classify, Detect Spans, Redact, Verify, and Extract Insights, which pulls out parties, dates, obligations, and risk. Detected items are labeled by type, not just flagged, e.g. tagging a company name as an organization rather than a generic match. Each stage runs on a distinct Featherless-hosted model sized to its job: Qwen3-0.6B classifies, Qwen3-8B detects spans, Qwen2.5-7B-Instruct verifies, Qwen2.5-14B-Instruct provides an independent second opinion, and Hermes-4-14B extracts insights and drafts NDAs. Every call is live to Featherless. Nothing is mocked. The standout feature is Model Comparison: a second independent Featherless model checks the same document and shows exactly where it agrees or disagrees with the first, with full reasoning, not just a score. Output is real and usable, not just an on-screen result: a downloadable, ready-to-send redacted document PDF, and a downloadable compliance summary PDF showing what was caught, confidence level, and by which model, an audit trail a lawyer could hand to a client. The market is hundreds of thousands of solo and small-firm US practitioners without an enterprise legal-tech budget, served via a low-price, per-seat or per-document subscription. Next up: batch processing, more contract types, HIPAA/GDPR/attorney-client compliance presets, and team accounts with shared audit history.