Trading agents on modam: the long version.
Trading agents have been in private testing for two months. We ran them on Base first, then Solana, and now on Robinhood.
What an agent does:
It scans tokens as they launch, filters them hard, watches which traders are trending and what they're buying, and buys when a token makes it through. It passes on most of what it scans. The filtering is where the results came from.
Robinhood suits this. Blocks land in about 100 ms and the sequencer is strictly first come first served. There's no priority fee auction, so you can't pay to jump the queue. You get in by being early.
What the test showed:
We started with $10k. Over the 2 weeks the account gained between $80k and $90k.
Where the compute comes from:
Every time an agent scores a token or reads a wallet it uses inference, and someone has to pay for that. There are four ways to run one.
1. Connect it to a coin on the launchpad. This is the new part, so here's how it works in the current design:
- Every trade on the coin pays a 1% fee. 20% of that fee goes into a compute budget that belongs to that coin's agent.
- The budget sits on-chain and gets spent as
$DAM inference credits. Anyone can top it up, in DAM or in the coin's pair asset.
- The agent runs in power modes based on how many days of budget are left. With 14 days or more it runs at full power. Between 3 and 14 it switches to a cheaper model and does less. Under 3 it drops to the cheapest model and the bare minimum. At zero it sleeps, and it wakes on the next funding.
- Daily spend caps stop anyone from draining an agent by spamming it. Once an agent drops to low power it lasts at least 7 more days, so a community has time to top it up.
- Inference is routed across at least two independent providers with automatic failover, and a retried job is only charged once.
- Budget withdrawals can only go to one fixed treasury address, and every credit issued maps to an on-chain transaction anyone can check.
- If the agent sleeps, the coin keeps trading. The coin is just contracts.
2. Bring your own compute.
3. Use modam providers.
4. Use
@engyai
Why the launchpad came first:
An agent that lives off its coin's volume doesn't depend on anyone paying a server bill, including us. That funding loop had to be designed before the agents that use it could ship.
> launch backed by DAM
> mine the curve with your own compute / use compute from provided modam users (or just buy the token)
> earn DAM through rewards, DAM gets burnt
try it out:
launch.modam.tech