Solving trust in agentic finance

Agents will manage your money. They’ll make mistakes we can’t predict - or even explain.

Meet Pakt: guardrails your agents can’t breach - even if they write faulty code, hallucinate or get hacked.

Live in private beta - secure agentic trading on Hyperliquid: usepakt.ai

Pakt in action - stopping GPT-6 Astra from losing me money on Hyperliquid:

Pakt: Let your agents trade. Stay in control.

Pakt puts guardrails between agents and your money. Set limits agents can’t breach - even if they write faulty code, hallucinate or get hacked. Any agent, any venue. Non-custodial. Private.

How it works

1. Pakt generates constraints for your strategy.
2. Your agents execute on full auto, proposing trades, self-optimizing and reacting to live data and events
3. Pakt verifies and only signs trades that are within your limits.

Under the hood: Fast ZK proofs, verified in TEEs

• Pakt requires agents to generate a ZK proof that every proposed trade complies with your limits. A trusted execution environment (TEE) verifies the proof and signs the trade for execution if - and only if - the proof is valid.
• Your credentials - private keys, exchange API keys and broker logins - stay sealed inside the TEE, inaccessible to anyone but you: not us, not your agent, not your cloud provider.
• Each trade comes with a verifiable receipt, letting you prove your PnL, trading volume and more without revealing individual trades.
• Proof generation takes 1.5 seconds on an RTX 4000 today, with up to 10 trades per proof. We’re targeting 0.5 seconds with further optimizations and faster hardware.

Try out the private beta: usepakt.ai

Interested in the tech? Check out this article by Dom, co-founder and mastermind behind Pakt, and the docs.

Our thesis: Agents will manage money. Trust is the #1 bottleneck.

Agents changed how we search, write code, communicate and design. They are changing how we handle our day to day tasks. Inevitably, agents will change how we manage finances. Agents are better at math, faster at learning, more efficient at processing data, have no emotional attachment to investments and they're online 24/7.

We believe that within 12 months, agents will be handling money for millions of people. And we're not the only ones betting on this. By 2030, McKinsey expects agents to mediate $3-5 trillion of global consumer commerce and BCG projects $370 billion+ additional annual profit for retail banks from AI by 2030.

But why is there no Instinct or Muse for trading and personal finance yet?

• It's still early. Frontier models have made huge leaps in the last few months and humans are playing catch-up. There are likely hundreds of teams working on this as we speak. Crypto was early to try agentic trading but struggled for this and other reasons - more on this below.
• Trust and safety. Using an agent to research an investment is very different from giving it permission to move your money. Agents can write buggy code, hallucinate, misunderstand instructions and get hacked. A single mistake can undo hundreds of good decisions. Better models and harnesses will reduce some of these failures - but they will never eliminate them.

Until we have reliable guardrails, “I don't trust agents with my money” will remain the default response of most users.

Gartner expects 40% of enterprises to demote or decommission autonomous agents by 2027 because of governance failures. They also project AI security spending to reach $4.8B in 2027 and $7.7B in 2028.

Our bet is simple. Adoption is inevitable, but it will require guardrails that enforce your limits. Forget “human in the loop.” The future is autonomous, and those who don't adapt will be left behind. If we can trust self-driving cars with our lives, why not trust agents with our money?

Guardrails will be to agentic finance what account abstraction is to DeFi and stablecoins are to payments: the unlock that makes mass adoption possible.

And this is why we are building Pakt.