How Legal Desk AI is built
We’re open about the things that matter for your trust: how we treat your data, and what we promise to keep it private. What we deliberately don’t publish is the internal blueprint. Naming every model, service, and layer would only help someone trying to attack or copy the platform, not the lawyers we build this for. So here’s what we stand behind.
Your data never trains AI
Client confidentiality is a professional obligation, not a feature. So the platform is built around a simple rule: your client data never trains AI. We do not use your prompts or the AI’s responses to train models, and we do not sell or share them.
When you use an AI feature, we temporarily store the prompt and response only to keep the service reliable and diagnose errors. Those logs are kept for a maximum of 28 days and then automatically deleted. Full details are in our Privacy Policy.
Private and isolated by default
Your legal work is yours. We keep it private and isolated to your account, and we never pool it with other users’ work. If an account stays inactive for 90 days with no pending balance, its legal data is deleted, so it doesn’t sit around forever.
Reliable, leading AI
We’re not a single-model wrapper. MAYA works across multiple leading AI models, so your tools stay fast and available even if one provider has an outage. Whichever model handles a request, the same rule holds: no provider receives your data for training, and your content stays inside your workflow.
Who builds it
Legal Desk AI is built by FuzzyCloud, a software studio founded by Kunjan Dalal, a software engineer since 2008 and an LL.B. graduate. An advisory board of practicing professionals helps guide it. The fuller story is on our About page.
More of our work
A few other things the team has built and looks after: