The agent
How it decides: one cycle, its tools, models, interval and critic.
OpenRouter · chosen at launch
What the agent is
A language model on OpenRouter, picked by the launcher and run under the protocol's account. It reads the thesis and the markets and proposes trades. It never holds a key and cannot place an order: code does that, inside the launch limits.
Every 1 min to every 24 h
One cycle
On its interval the agent runs one decision cycle: 9 steps, 2 of them the model's. Each step shows on the coin's live console as it happens. A cycle can end with no trade at all.
Budget gateCode
Checks the cycle's estimated cost against the budget and today's cap. Short: the agent sleeps.
Market scanCode
Picks the markets the launch allows that clear the liquidity floors: at most 40, the most liquid first, plus anything it holds.
SnapshotCode
Freezes prices, funding, the account and open positions into one snapshot, hashed before the model sees it.
Model (OpenRouter)Model
Reads the thesis and the snapshot as data, may call read-only tools, and answers in a strict format: the whole target book and a short public summary.
Critic (optional)Model
A second pass that can veto the decision, never change it. If it cannot run, nothing trades.
Risk engineCode
Checks every target against the launch limits and the live market, then approves, shrinks or rejects it, and records why.
Order plannerCode
Turns approved targets into orders: exits first, sizes and prices rounded to the venue's rules, limit orders within the maximum slippage.
Execution, exactly onceCode
Sends each order exactly once: on Pacifica for perps, as a co-signed vault swap through Jupiter for spot. In PAPER, fills are simulated against the live order book.
NAV, high-water mark, buybackCode
Books the fills, marks the treasury, and queues the buyback share of any profit above the high-water mark.
Every step in full, with what stops it and where it shows: How it works.
Set at launch
What the launcher sets
Model
- Choice
- Any model the launch lists from OpenRouter's catalogue.
- Free models
- No cost, but 50 requests a day (1,000 once $10 of credits has been bought), and their provider may log prompts. They are badged.
Decision interval
- Range
- Every 1 min to every 24 h, 1 h by default.
- Trade-off
- Shorter intervals mean more model calls, more inference spend and more trading fees. The launch shows calls and cost per day.
Tool calls
- Range
- 0 to 8 per cycle, 4 by default. 0 means a single answer with no lookups.
- Tools
- Read-only market data on markets in the snapshot: candles, order book, funding history. None of them can write anything.
Reasoning
- Levels
- Low, Medium, High; Medium by default. More reasoning costs more per call.
Critic
- What
- A second model pass that can block a decision. On by default; it costs one more call per cycle.
Pacifica · Solana
Where it trades
The launcher picks one or both. A market id says where it trades: PAC:NVDA is a Pacifica perp, SOL:NVDA_XSTOCK a token the vault holds on Solana.
Pacifica perpsProtocol custody
- Markets
- A perpetuals exchange on Solana with an off-chain order book, margined in USDC: crypto, single stocks (NVDA, TSLA, META and others), the S&P 500, gold, silver, oil and FX (EUR/USD, USD/JPY).
- Trades
- Long or short, with leverage up to the launch setting (at most 10×).
- Note
- Stock, index, commodity and FX perps trade around the clock on an oracle price and are thin: tens to hundreds of thousands of dollars a day.
Solana spotProgram custody
- Markets
- The vault itself swaps through Jupiter: verified Solana tokens and xStocks (tokenized stocks).
- Trades
- Long only, no leverage.
- Note
- xStocks are restricted by jurisdiction: switched on per deployment, and never offered where they are not permitted.
Every market must clear the protocol's liquidity floors, which no launch can lower: at least $25,000 of 24h volume and, for perps, $15,000 of open interest. They are low enough to admit the thin perps, so the risk engine also caps every position at 1% of the market's 24h volume and of its open interest.
Public
Watch it think
The console shows the model's short public summary and its rationale per position, never its prompts or raw reasoning: the model writes the summary for readers. Every decision record carries its input snapshot, and the page recomputes the snapshot's hash in your browser.
It decides from market data and the thesis only. Signal sources such as X accounts and Telegram channels are SOON. Open a coin to watch its agent.