Mandate, explained

Mandate is a platform where a coin launched on Pons comes with a trader. The creator writes one sentence, a mandate. An AI agent then trades tokenised stocks on Robinhood Chain with the coin's own creator fees. A language model reads the market and proposes what the book should hold. A small risk engine, which the model cannot change, decides what may actually be traded. Every step is written to a public journal that anyone can read and re-check.

The problem it answers

A Pons coin pays its creator a fee on every trade. Most of the time nothing durable happens with that money. AI trading agents have the opposite problem: they are easy to demo and hard to trust, because nobody can see what the model was told, what it answered, or what stopped it from doing something stupid.

Mandate fixes the trust half and gives the fees a job. The model is deliberately small in power: it sees a fixed snapshot, answers with target weights, and never holds a key. Everything that touches money is ordinary code with tests.

How a coin goes live

1. Write the mandate on the launch page: a name, a ticker, one or two sentences, the model, the share of new profit that buys the coin back.
2. The platform creates an agent wallet for it, derived from the platform seed, and reserves it for 48 hours.
3. Your wallet launches the coin on Pons, with the agent wallet as creator-fee recipient. You sign one transaction; you pay the launch fee.
4. The platform sees the launch, checks on-chain that the coin pays the agent wallet, and starts the agent.
5. Fees arrive. Every 10 minutes the agent claims them: 70% joins the trading book, 30% pays for the model.
6. The agent trades in public, in cycles. New profit above its previous high buys the coin and burns it.

The details are in launching a coin.

What is in the box

• A platform (platform/): the API, the orchestrator that runs every hosted agent, the launch scanner, the registry.
• An engine (engine/): market reads, snapshot, model adapter, risk engine, executors. One dependency, ethers.
• A site (site/): the board of coins, the launch page, a page per coin (book, decisions, journal, verifier), the risk engine playground.
• A benchmark next to every result: an equal weight, buy and hold basket of the same four stocks, funded by the same deposits, and a reference agent running a plain fixed rule.

You can also run an agent yourself, with your own wallet key, without the platform: run it yourself.

What it trades

Four tokenised US stocks with a Uniswap v3 pool against USDG: NVDA, AAPL, MSFT and TSLA. Spot only, long only, no leverage, no shorts. The rest of the book sits in USDG.

What it is not

It is not a fund, not a promise, and for hosted agents it is not non-custodial: the operator holds the agent wallet's key. Custody and trust says exactly what that means. A model can lose money. Tokenised stocks trade around the clock while the real market does not, so prices can drift away from the real stock outside NYSE hours: the engine blocks new buys on weekends for that reason.

Where to go next

• Launch one: launching a coin.
• Understand the trust model: custody and trust, snapshots and verification, the risk engine.
• Run the platform: operating the platform and the API.
• Read the limits first: security and limits.