Hey everyone,

We’re Vrushank and Kenny, co-founders of Sanctum.

TL;DR

Sanctum gives you a preview of how real users will react to your next release. We simulate your customer base with AI user models, find what breaks, show where users get stuck, and help you choose the best product direction before shipping.

https://youtu.be/_0zi_Olj2r4

Problem

Software cycles are faster than ever, but the feedback loop hasn’t kept up. Teams still learn what worked only after a launch, when fixes are costly and the damage is done. Testing is tedious and fragile, and even with AI involved, it rarely reflects how real users actually behave. Canary deploys and feature flags offer some protection if you have scale, but they still put real users at risk. A/B tests add time and cost, so teams narrow their experimentation and overlook better ideas.

The result: teams are still building for users they don’t truly understand.

Solution

We create user models that behave like your real customers, built from session recordings and a mix of rich synthetic data and real usage patterns. Our methods are grounded in recent research on audience-aligned user behavior simulation and our own modeling techniques based on alternative data. We then unleash the agents on your new product changes, like an artificial canary deployment.

If your apps are containerized, we can spin up your full stack with forked DBs and dependencies inside ephemeral cloud environments to run simulations with integrated telemetry for bug diagnosis and profiling via our SDK. If you aren’t containerized, we can still run the simulations by attaching to your dev or staging URLs. Results can optionally be fed back to AI tools via MCP.

You get a preview of how real users will react, where they stumble, and what breaks.

Get artificial users built from real, anonymized user profiles endowed with your session recordings and customer data to interact with your product and answer your questions. We can scale to thousands of parallel simulations at a time and use SSR to provide meaningful metrics.

Whenever you make a PR, our GitHub action will trigger a “simulated prod” canary deployment and surface recordings of discovered bugs from smoke tests and diff-inferred test paths. This can also be triggered via CLI. Our SDK which wraps OpenTelemetry can help diagnose these issues.

Vision

As agents continue to improve, their interactions with software will become indiscernible from human behavior. Our agents will be able to interact with your application in any form: web, mobile, video, voice, and more.

Building user models based on your target audience gives product and engineering teams a powerful tool to test and learn at scale. While it’s not a replacement for talking to customers, it allows you to explore a larger idea space, identify promising directions earlier, and iterate faster.

As agent traffic grows, we aim to make agent usability and UX feedback a signal for developers. We’re building the devtool for engineers and product teams that closes the loop on feature development: use AI tools to build your software, and use us to evaluate it.

Our Story

We just graduated from Stanford (BS + MS, CS) and both spent time in quantitative finance, where data is the answer to every question. We’ve known each other since the first days of college, and have been close friends and collaborators ever since.

Collectively, we’ve conducted cutting-edge AI and cryptography research, constructed highly profitable trading strategies, and built core engineering infrastructure for some of the most sophisticated institutions in the world.

We believe that data-rich simulation environments are the next frontier in analytics. Our goal is to build them with the right safeguards and privacy standards from day one. We ultimately want to help organizations build better solutions for humans.

Ask

If you:

• are interested
• use PostHog, Amplitude, or session replay tooling
• use LaunchDarkly or feature flags
• wish you could ask your users more questions, more often
• or have ever shipped something that seemed fine... until users touched it

We would love to hear from you! Email founders@heysanctum.com or book a call / express interest at heysanctum.com