AI trading agents, native to Base.
A suite of autonomous agents, each focused on a single on-chain strategy and driven by an LLM. A coordinator splits capital across them and watches the overall risk.
- 6 + 1agents + coordinator
- 3modes: paper · dry-run · live
- MITopen source
Experimental software, not financial advice. These bots can trade real money on Base and may lose some or all of the capital you give them. Start with paper trading or dry-run; when you go live, use dedicated wallets and only amounts you can afford to lose.
The model proposes. The executor decides.
Every agent runs the same decision cycle. Risk limits live in the executor's code and cannot be bypassed from the prompt.
- Market dataDiscovery of prices, yields and opportunities.
- Portfolio stateBalances and open positions, read from the chain.
- Automatic exitsRisk exits run before the model is even consulted.
- LLM decisionThe model answers with structured JSON.
- Executor checksLimits are enforced in code, then the order executes.
One agent, one strategy.
Each agent lives in its own repository with its own deployment, and can be used on its own.
perp
Directional perpetuals with leverage.
degen
Speculative spot on new tokens and memecoins, with security screening.
yield
Passive yield on lending markets and vaults, with optional looping.
neutral
Delta-neutral funding carry: long spot plus short perp.
dca
DCA scaled by the Fear & Greed index, plus rebalancing.
lp
Concentrated liquidity with range re-centering.
coordinator
The control room: market regime, capital allocation, circuit breaker and gas. It reads and commands the agents through their APIs.
Built to fail small.
An LLM can be wrong. The executor's limits reduce the damage; they do not eliminate it.
Three modes
PAPER_TRADING uses a virtual portfolio with real prices. DRY_RUN reads the chain and prints the plan, signing nothing. Then live. The dashboard always shows the active mode.
Protected commands
Every endpoint that acts stays disabled until DASHBOARD_RUN_TOKEN is set.
Dedicated wallets
One wallet per agent, small amounts, and never reuse the private key elsewhere.
Telegram reports
A report after every cycle, error alerts, and commands accepted only from the configured chat.
Paper first. Live later.
- Pick an agentFollow its README;
.env.examplelists every variable. - Run in paperLet it work for a few days and check P&L, costs and decisions.
- Go live, smallUse a dedicated wallet and an amount you can afford to lose.
- Add the coordinatorOne dashboard, capital moved between agents.
Note: the individual agents' READMEs are written in Italian for now.