Everything local.
Nothing leaked.
Vantom’s pipeline runs entirely on your machine — market data in, structured verdict out, zero telemetry, zero round-trips to a server.
Data pipeline — 5 stages
Live Market Data
3 sourcesBinance WebSocket · Alpaca · Finnhub — asset class auto-detected
Detection Engine
per-candle scanScans every candle for FVGs, Sweeps, and BOS in real time
Confluence Scorer
signal stackingSignals that cluster in the same zone stack their scores
AI Gate
thresholdOnly setups above the confidence threshold reach the model
Verdict
local LLMStructured JSON — entry, target, invalidation, thesis — on-device
Design Principles
Local-first by design
The reasoning model runs on your Mac. No API keys. No per-call cost. No position data transmitted anywhere. The verdict never touches a server.
Asset-class-aware
Crypto, equities, and forex behave differently. The detection engine uses separate parameter profiles per asset class so a BTC signal and an AAPL signal are evaluated on their own terms — not the same ruler.
Parallelism without blocking
Market data processing and UI updates run on the main thread. AI inference runs in an isolated background process. A slow model call never freezes a chart or drops a candle.
Deduplication built in
The same setup doesn't fire the AI twice. A fingerprinting layer tracks active setups and suppresses redundant calls — so you get one clean verdict per opportunity, not a flood.