Polarity, Inc.
Polarity Labs builds the most accurate eval infrastructure for AI agents through sandboxed environments that mirror production conditions.
At a Glance
- AI development teams
- Enterprise engineering teams
- Companies shipping autonomous agents
AI Tools by Polarity, Inc.
(1)
ZenithDB
Open Source AI Agent Trace DB
Discussions
No discussions yet
Be the first to start a discussion about Polarity, Inc.
Latest News
Products & Services
Sandboxed eval infrastructure for AI agents that runs tasks in isolated Docker containers with real backing services.
Described as a 'world-class AI QA engineer' designed to catch bugs and validate agent behavior.
Market Position
Differentiated from Braintrust, LangSmith, and Langfuse by using real-service sandboxes per run rather than mocked dependencies, targeting complex multi-step agents.
Leadership
Founders
Shane Barakat
Founder @ Polarity, Engineering @ University of Waterloo. Background in engineering and building AI agent infrastructure.
Alex Ungureanu
Founder @ Polarity, CS & Math @ University of Waterloo. Focuses on agent direction, specs, and research.
Executive Team
Shane Barakat
Founder
Engineering lead and founder.
Alex Ungureanu
Founder
Research and product lead founder.
Founding Story
Polarity was started to address the gap where AI agents fail in production despite passing prompt-level tests. The founders built Keystone to provide run-level testing in production-shaped sandboxes.
Business Model
Revenue Model
Subscription-based SaaS with additional usage-based compute billing ($3-$5 per GB after allowance).
Pricing Tiers
1 GB processed data, 20 concurrent sandboxes, 7-day trace retention, unlimited projects & evals.
5 GB processed data, 1,000 concurrent sandboxes, 30-day trace retention, custom evals & environments, SOC 2/GDPR/HIPAA.
Unlimited concurrent sandboxes, volume discounts, BYO cloud/on-prem, SSO + SCIM, Dedicated solutions engineer.
Target Markets
- AI development teams
- Enterprise engineering teams
- Companies shipping autonomous agents
- Production AI agents
- Long-running multi-step agents
- Stateful behavior validation
- Regression testing
- Agent debugging
- Composio
- Cal.com
- StackAdapt
- Stan