We help you build AI trading agents using our real-time market data, causal analysis and pre-built agent framework. Start from our four agents or wire the API into your own, and test on paper before you risk SOL.
Waiting for market feedSetup score = Σ weight × component (0–1) / Σ weights, missing data scores 0, bar activity excluded until kline loads, × 0.4 if collapsing (1h < −50%). Gates: liquidity ≥ $15.00K, top 10 ≤ 40%, safety ≥ 40, dev ≤ 10%, known security.
No open paper positions. Rules: TP1 sells 50% at 1.5R, TP2 sells the rest at 3R, stop to break-even after +1R, trail 1R below the peak after TP1, max hold 45m, escape at severity ≥ 50.
The first report is built after 5 closed trades (or 90 ticks after the first). Proposals need at least 30 trades per bucket and a 10-point effect, move 20% toward owner bounds, and never touch caps or the kill switch.
Hover or focus a node. Solid arrows are relationships the causal search kept this window; the dashed arrow between pools only lights up when the CEDAR edge passes alpha 0.01 on all seeds.
Every number on this page comes from live market data or a recorded replay of it, run through the same signals and arithmetic we expose in the ARCTIS API.
A causal worker models the ARCTIS pools with Bloomberg's causal-ts methods. CDNOTS+ detects regime breakpoints (high or low activity) through a time-index node; CEDAR tests lead-lag between pools and only reports an edge that persists at alpha 0.01 across 5 seeds.
Kronos, a foundation model for K-line data, reads recent OHLCV candles of the ARCTIS pool and forecasts the next close, high and low. The live market maker uses it as a bounded quote skew. In this demo it is shown next to the chart as context; it does not trigger entries until backtests show it adds edge.
Loading forecasts.Pool data comes from the ARCTIS PumpSwap pool and its Raydium CPMM pair: prices, spread and volume per minute. The causal worker and Kronos both read these pools, and the lead-lag strip under the chart shows which venue currently moves first.
The agents use only public building blocks: 23 REST endpoints, an MCP server with 23 tools, 15 agent skills and 5 prompts. Anything this demo does, your own agent can do with an API key.
Setup score is a weighted sum of liquidity, safety, holder spread, bar activity, turnover and buy ratio. Entries need a 15-minute flush; the stop is the tighter of a fixed percent and 4x Wilder ATR 14, with VWAP 12 and the 24-bar swing low shown for context. TP1 and TP2 sit at 1.5R and 3R, size is capped by risk per trade, and fees plus modeled price impact are charged on every fill.
We replay recorded market data, compare strategy versions and tune parameters within fixed bounds. The Reporting agent proposes the same kind of bounded changes live. When the results hold up out of sample, we will release the strategy publicly. Until then this stays a paper-trading beta.
Closed paper trades appear here with fees (50 bps per side) and modeled price impact included.
Waiting for market feed