Nani docs
Nani is a crypto wallet for macOS with an AI assistant built in. Hold your ETH and tokens, talk to the chain in plain English, and approve everything with Touch ID.
What you can do
- Hold and send ETH and ERC-20 tokens on Ethereum, Base, Arbitrum, Optimism, and Sepolia.
- Ask the AI for anything on-chain — prices, ENS names, balances, swap quotes, gas, address risk checks.
- Send by talking. "Send 0.05 ETH to vitalik.eth" produces a transaction you review and approve.
- Run the AI on your Mac. A small fine-tuned model runs locally on Apple Silicon. No cloud needed, no API key needed, no usage limits.
- Switch to a bigger model when you want one. OpenRouter and Ollama are one click away in Settings.
Why local-first
Most wallets and most AI assistants send your data somewhere. Nani is built the other way around:
- The wallet keeps your private key in the macOS Keychain behind Touch ID. Only the wallet can read it, and only when you approve.
- The default AI runs on your Mac. Your prompts never leave the machine.
- There is no analytics, no telemetry, no sign-in, no account.
You stay in control of two things people normally hand over: your money and your conversations.
How it fits together
Three parts: the wallet, the assistant, and the tools the assistant uses.
- The wallet owns your private key. It's the only thing that can sign.
- The assistant is a language model. It can talk and it can call tools.
- The tools read the chain and propose actions. When the assistant proposes sending money, the wallet shows you the parsed transaction and asks for Touch ID. The AI never touches the key.
This separation is what makes giving an AI access to your wallet safe.
Where to next
New here? Start with Install, then Set up your wallet.
Already running Nani? See what the AI can do in AI assistant and Tools.
Want the deep dive? Security covers how keys are protected. FAQ answers the common stuff.
Open source
Nani is open source. The macOS app is at github.com/NaniDAO/nani-macos, the AI tools are at agentek, and the model is on HuggingFace. Join the waitlist for the public build.