The evolution of Mizuki. A private AI that verifies on-chain activity with transparent, reproducible evidence. Public Robinhood chain auditor.
Mizuki exposed bad actors through commentary. Mizukara does it with verifiable analysis. The goal is not to be believed, but to make trust unnecessary.
Your data stays local. Files, memory, and conversations never leave your machine.
Every conclusion is backed by evidence, not just a verdict.
Every analysis is reproducible. Re-run it and check the result yourself.
Don't trust the AI; inspect the proof. MIT licensed, source public.
Everything runs through one local gateway. Only the model call can leave, and only if you choose a remote provider.
The public instance analyzes on-chain activity and publishes every finding with the evidence behind it.
Contracts, wallets, permissions, and deployer history, read straight from the chain.
Every finding links to a replay log anyone can re-run.
Priority monitoring and full access on the public instance.
# 1: install the runtime
uv tool install "mizukara[recommended]"# 2: connect a provider, or point it at ollama and stay local
mizukara onboard# 3: start the gateway
mizukara gateway run