The Story Behind Mentra

Inspiration

It started with a moment that felt wrong.

Watching a colleague paste an entire internal project brief — client names, financials, strategy — into ChatGPT to get a summary faster. No hesitation. No awareness of what was leaving the building. That moment made one thing clear: the problem wasn't the employee. It was the absence of a better alternative inside the company.

If the tools available internally were as powerful as the public ones, nobody would need to go outside.

What We Learned

Discovery interviews with 10–20 CTOs and CISOs confirmed it wasn't an edge case. Every security leader we spoke to had the same story — employees defaulting to public AI because internal tools couldn't compete. Banning didn't work. Education didn't work. The only real fix was building something better that stayed inside.

How We Built It

We designed Mentra around one principle: AI should be embedded at the task level, not bolted on at the org level. Every task carries its own context — documents, goals, constraints — so the AI never needs to reach outside to answer a question. We built on a tenant-isolated architecture using RAG scoped per task and LoRA adapters deployed inside the customer's own environment.

The stack: Supabase for auth and data, edge functions for real-time AI inference, and a Vite + React frontend designed to feel familiar to anyone who's used Monday.com — but faster, cleaner, and actually secure.

Challenges

The hardest part wasn't the AI. It was scoping it correctly. Making the model only answer from task-level context — not hallucinate, not pull from the internet, not bleed across tenants — required architectural decisions that most AI productivity tools skip entirely because they're hard.

We didn't skip them. That's the product.

Built With

• 4.6extended
• antigravity-cloud:-aws-/-azure-/-gcp-(vpc-deployment-model)-security:-byok-encryption
• auth)-ai/ml:-anthropic-claude-api-(claude-sonnet)
• languages:-javascript
• lora-adapters-edge-functions:-supabase-edge-functions-(deno-runtime)-web-search:-tavily-/-serper-dev-tools:-cursor-pro
• rag-architecture
• rbac
• row-level-security
• sonnet
• typescript-frameworks:-react
• vite-styling:-tailwind-css-v4-backend-&-database:-supabase-(postgresql