Enso is Hanzo AI's frontier model family and the router behind auto: every request is classified, priced against what your org can serve, and answered by the model that should answer it. Three tiers keep their order on quality and price as the models under them change.

Model	For
`enso-ultra`	the hardest problems, top-tier accuracy
`enso-pro`	everyday work at frontier quality
`enso-flash`	high volume, low latency, low cost
`enso-auto`	let the router pick for each request

Enso is a Hanzo AI original: proprietary and closed source. There is no code or weights here; it is served only through the Hanzo API.

Model Card: Architecture, Benchmarks & Economics · Hugging Face Hub
Benchmarks, measured on one harness: hanzo.ai/models/enso
Docs: docs.hanzo.ai/docs/models/enso

Use Enso through the Hanzo API

curl https://api.hanzo.ai/v1/chat/completions \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "enso-auto", "messages": [{"role": "user", "content": "Hello"}]}'

The API speaks the OpenAI wire format, so any OpenAI SDK works with base_url="https://api.hanzo.ai/v1".

Examples

Runnable client examples in this repository:

Python: examples/python/client.py

pip install -r examples/python/requirements.txt
export HANZO_API_KEY="your-api-key"
python examples/python/client.py

TypeScript: examples/typescript/client.ts

cd examples/typescript
npm install
export HANZO_API_KEY="your-api-key"
npm start

Go: examples/go/main.go

export HANZO_API_KEY="your-api-key"
go run examples/go/main.go

Rust: examples/rust/

export HANZO_API_KEY="your-api-key"
cargo run --manifest-path examples/rust/Cargo.toml

cURL: examples/curl/chat.sh

export HANZO_API_KEY="your-api-key"
./examples/curl/chat.sh

Production Rust Recipes

recipes/: Modular, typed Rust patterns for model routing:

ModelSuitabilityRecipe: Single-shot classification predicting optimal model capability tier given prompt tokens, domain, and latency/cost bounds.
ConfidenceGate: Evaluates Kai decision margins (\Delta = P_{(1)} - P_{(2)}) against threshold \tau for low-latency fast decisions vs. LLM escalation.