About Kamiwaza

Building the routing layer enterprise AI has been missing

Founded in San Francisco in 2024. Single founder. Bootstrapped. Built after watching the same infrastructure gap appear at every large organization that tried to deploy LLMs at scale.

Why we built this

Why we built this

Enterprise AI adoption stalls not because models are inadequate, but because the infrastructure to deploy them responsibly — tenant isolation, data class governance, and a complete audit trail — doesn't exist as a coherent, off-the-shelf product. Every organization re-solves it with custom glue code. An AI-native fintech routing 80M requests per month and a regional health-tech platform managing 12M monthly inferences are solving the same three problems independently. Kamiwaza replaces that glue code with a declarative routing layer.

Kamiwaza is not a model provider. We route to your existing keys. We do not mark up tokens. We do not store prompt text. We make sure the right prompt reaches the right endpoint, under the right policy, with a complete audit record.

Small technical team working at laptops in a modern San Francisco office with monitors showing code and data
The team

The team

Luke Norris, founder of Kamiwaza

Luke Norris

Founder & CEO

Luke started Kamiwaza in San Francisco in 2024 after leading platform engineering at Faction, where he built the data-plane routing infrastructure for a multi-tenant cloud storage product. The job required handling tenant key isolation, per-request policy evaluation, and structured audit logs at scale — the same three problems he kept seeing platform teams struggle with when they started deploying LLMs. He left Faction in early 2024 to build Kamiwaza as the product he wished had existed.

Kamiwaza is independently funded. There are no outside investors.

Principles

How we build

Audit trail first If it can't be logged, it shouldn't be routable. Every policy decision leaves a structured record you can query, export, and use in compliance audits.
Configuration over magic Routing behavior must be explicitly declared in YAML. No implicit rules, no "smart" defaults that surprise you in production. What's in the config file is what happens.
Privacy by architecture Data class isolation is structural, not optional. PII staying on-prem isn't a setting you toggle — it's a constraint enforced by the routing engine at request evaluation time.