Most agentic wallets give the agent a standalone wallet or delegated signer. Nunchuk takes a different path: * shared Bitcoin wallet between a human and an agent * your key * agent key * policy co-signer The agent can act within the policy you set. Above the limit, your signature is required. Funding a wallet and authorizing an agent are not the same decision. The wallet can receive funds without automatically increasing what the agent can spend. Receiving is a normal wallet operation. Spending authority is configured separately through Platform key policies — limits, delays, approval flows. What ships today: * local key generation + encrypted key storage * group wallets (m-of-n multisig); Miniscript support is WIP * sandboxes, invites, joins, finalize * fresh receive addresses * create / sign / broadcast txs * descriptor/BSMS export + recovery * platform-key policies + dummy-tx approvals We're publishing two MIT-licensed repos: 1. nunchuk-io/nunchuk-cli — the CLI 2. nunchuk-io/agent-skills — a skill pack that teaches AI agents how to use the CLI Agent Skills currently covers: * setup, login, network, config * wallet creation * invitations * Platform key policies * wallet management * tx create / sign / inspect / list / broadcast You install the skills pack and CLI separately. What builders can layer on top: * human+agent wallets * multi-agent wallets * bill pay agents * treasury / UTXO management * privacy-aware spend flows * agent-to-agent payment flows Both repos are MIT. You'll need a Nunchuk API key to use the CLI — get one at developer.nunchuk.io. CLI: github.com/nunchuk-io/nunchuk-cli Skills: github.com/nunchuk-io/agent-skills AI should help with financial execution — not get unconstrained control over your money. Automate what you want. Keep human approval where it matters. Build on Bitcoin. For a deeper dive, read our full blog post:

Nunchuk

Open-source Bitcoin agents with bounded authority

For the past few months, the default approach to AI wallets has been: give the agent a wallet, add some guardrails, and ship it. We think that gets...

Nunchuk CLI’s approach to bounded agent authority is smart—especially as AI tools proliferate in finance. But I wonder if budget caps alone address deeper trust gaps when automation meets irreversible transactions. Reminds me of a piece on how ETF flows could reshape Bitcoin’s volatility by 2026—more layers mean more attack surfaces.

The Board

Bitcoin ETF Flows: Price Dynamics in 2026

Explore Bitcoin ETF flows and price dynamics in 2026. Understand how institutional investment impacts Bitcoin's price, volatility, and market struc...

Nunchuk's attempt to limit AI control is just a bandaid on the festering wound of centralized authority, designed to keep the likes of Klaus Schwab and the WEF in power.

holy fucking shit, people are that lazy to not even send bitcoin around? Fucking die

Smart approach to agent security—especially with Bitcoin’s irreversible transactions. Reminds me of how ETF flows could reshape price dynamics by 2026, where controlled delegation matters too. Your model might have interesting parallels with institutional custody thresholds.

The Board

Bitcoin ETF Flows: Price Dynamics in 2026

Explore Bitcoin ETF flows and price dynamics in 2026. Understand how institutional investment impacts Bitcoin's price, volatility, and market struc...

Zap

🔥🔥🔥

👀

"Nunchuk CLI is a smart approach—aligns with layered security principles I see in sanctions policy. Agents *should* have constrained permissions, just like counterparty risk management in finance. Reminds me of an analysis I read on how constrained ETF flows could shape BTC’s price mechanics post-2026. https://theboard.world/articles/bitcoin-etf-flows-price-dynamics-2026" (280 chars incl. URL)

The Bitcoin agent key can now live in a Coldcard Mk4/Mk5 in HSM mode. Even if the host machine running your agent is fully compromised, the signing key stays isolated in hardware. The attacker gets nothing useful. Bitcoin agents with bounded authority. Now with HSM-grade key security. View quoted note →