Perceptron ML

Perceptron ML is best understood as a build partner, not a subscription. You do not log in to a product; you scope a system, and the vendor builds it around how your firm actually works. The four documented build types (timekeeping, client acquisition and monitoring, grounded research, discovery and drafting) are the workflows where mid-to-large firms lose the most partner hours. The grounding engine is the real differentiator. Every fact is derived from a primary source and checked before a model is allowed to use it, and the homepage proves it with a motion to exclude expert testimony where Daubert v. Merrell Dow Pharms., 509 U.S. 579, 589 (1993), Kumho Tire Co. v. Carmichael, 526 U.S. 137, 141 (1999), and Gen. Elec. Co. v. Joiner, 522 U.S. 136, 146 (1997) are each reported as reporter-checked or quotation-matched. That is a concrete claim a partner can independently verify, which is rare in legal AI marketing. The confidentiality posture is the second strength: deployed in your environment, work product never trains anyone else's model, and every answer traces to an openable source. For a firm whose outside counsel guidelines forbid sending client data to a shared model, that removes the standard objection. The honest weaknesses are the mirror image of the custom model. There is no published tier list, no self-service path, and no public documentation hub beyond the marketing site, so technical due diligence depends on the scoping conversation. No pre-built integrations with common legal software (Clio, NetDocuments, and the like) are documented on the pages we reached, so connectors are likely part of the build rather than a switch you flip. On-prem also means your firm manages the infrastructure. Where it fits: firms with 20+ attorneys, a defined workflow to automate, and a confidentiality mandate strong enough to justify custom work. Where it doesn't: solos and small firms who need a tool next week, and lean teams without someone to own an internal deployment.