Hoshang Mehta

Hoshang Mehta

PylarPylar
Co-founder at Pylar
631 points

Badges

Buddy System
Buddy System
Bright Idea 💡
Bright Idea 💡

5

Top 5 Launch
Top 5 Launch
Tastemaker
Tastemaker
View all badges

Maker History

Forums

Architecture design to keep AI Agents from compromising customer data

Most AI agents need access to structured data (CRMs, databases, warehouses), but giving them database access is a security nightmare. Having worked with companies on deploying agents in production environments, I'm sharing an architecture overview of what's been most useful- hope this helps!

Now Live on Pylar : Agent Data Connectors

One of the biggest blockers to building agents is getting the data 'agent-ready'. Teams spend months building pipelines, wiring up sources, cleaning data, and centralizing it - before an agent can even ask its first question.
Pylar now does this out of the box.
We re source-agnostic. Whether your data lives across multiple databases and warehouses (Supabase, Snowflake, MySQL, etc.), you can connect one or many instantly, no re-architecture required.
If you don't have a warehouse yet, we ve got you covered. Pylar ships with 100+ built-in integrations across marketing tools, CRMs, support platforms, product databases, and billing systems. Data comes in cleaned, transformed, and centralized, ready for agents to work with.
Next up is agent views - once you've connected to your sources, you can write SQL across or within to create precise, sanitized, sandboxed views purpose built for specific agents.
Agents don t roam your databases arbitrarily. You deterministically scope exactly what fields they can access, so they do their job well, without hallucinating or giving you different answers for the same/similar questions.
Give it a try and let me know what you think!

Launching Airbook AI - Cursor for Analytics

Today, we re excited to launch Airbook AI - an AI-native workspace to do all your analytics in one place.

Airbook automatically syncs, cleans, and centralizes data from HubSpot, Amplitude, GA4, Zendesk, Stripe, Postgres, Snowflake & more.

Our AI is schema-aware - it understands your question and maps it to the fields that exist in your database, auto-builds queries across sources, and lets you edit everything it generates too- so you're always in control. From there, turn results into dashboards via a prompt or trigger workflows to push data into other growth tools.

Three ways to work:

View more