Our Values

Ship With Confidence

We don’t ship fast and break things. We ship fast and know things. Every release should be backed by data - evaluations run, metrics checked, guardrails tested. Speed without confidence is recklessness. Confidence without speed is irrelevance.

Default to Transparency

We share context broadly. Internal docs are open to the whole team. Decisions are documented with their reasoning. Metrics dashboards are visible to everyone. When in doubt, share more, not less. This handbook itself is a reflection of this value.

Measure Before You Claim

Opinions are welcome. Unsubstantiated claims are not. If you say something is “better,” show the benchmark. If you say a feature “works,” show the test results. We build an evaluation platform - we should be the most rigorous evaluators of our own work.

Protect the User

Our customers’ users trust AI agents with real decisions - medical information, financial advice, customer support. Every hallucination we prevent protects a real person from a real consequence. This responsibility shapes our priorities.

Think in Systems

Individual components are simple. Systems are complex. We think about how pieces interact, where feedback loops exist, and what happens at scale. A guardrail that works at 10 requests per second but fails at 10,000 isn’t a guardrail - it’s a demo.

Stay Curious

AI is moving faster than any technology in history. What’s state-of-the-art today is outdated in six months. We invest time in learning, reading papers, experimenting with new approaches, and questioning our own assumptions. The moment we stop learning, we start falling behind.