Hanzo AI
Hanzo AI provides an AI acceleration platform that makes frontier models up to 200× faster and 400× cheaper for coding agents. Its Gen harness integrates the Enso router and Kai decider into tools like Claude Code and Codex, routing each call to the most cost-effective model. The platform installs via a single CLI command and supports self-hosted or air-gapped deployments.
- Artificial Intelligence
- Developer Tools
- AI Agents
- Software Only
Funding
Funding not disclosed
Founders
Founder details are not available yet.
Product
Problem
Frontier AI models are expensive and slow, and coding agents pay premium prices on every step—even for trivial operations like reading a file or deciding whether a test passed. Teams face an unavoidable trade-off between cost, speed, and quality, forcing them to either overspend on frontier models or settle for weaker performance.
Solution
Hanzo provides an AI acceleration platform that makes frontier AI models up to 200× faster and 400× cheaper. Its Gen harness weaves the Enso router and Kai decider into popular coding agents like Claude Code, Codex, Cursor, or its own Dev agent, sending each model call to the model that answers it right for the least. Enso routes requests dynamically across Flash, Pro, and Ultra tiers, while Kai answers agent choices as decisions in sub-millisecond latency on GPU, writing no text so no output is billed. The solution installs via a simple CLI command and works with the coding agents teams already use, moving the cost-performance curve without changing workflows.
Target Audience
Primary customers are software development teams and enterprises using coding agents who want to reduce compute costs and latency without sacrificing model quality, including organizations needing self-hosted or offline AI infrastructure.
Features
- Enso router trained on outcomes, sending each request to the model that answers it right for the least, with static routing or custom training on your own traffic
- Kai decider answering agent choices with probabilities in 0.27 ms p50 latency on GPU, writing no text so no output is billed, at $0.021 per million input tokens
- Gen harness that launches Claude Code, Codex, Cursor, or Dev with the Hanzo MCP toolset attached, linking sessions to the Hanzo account and routing model calls through api.hanzo.ai
- One-click CLI installation via `curl -fsSL https://hanzo.sh | sh`
- Full-stack platform including agents, models, memory, identity, machines, and a Record layer for audit trails
- Enterprise deployment options including running in your own cloud, your own servers, or fully offline