Barfinex — AI-Native Trading Operating System
Barfinex is a complete operating system for AI-driven trading. Five integrated services cover the full trading lifecycle — from market data ingestion to AI decision-making, risk governance, and execution.
Barfinex is not a trading bot. It is not a strategy framework. It is an operating system for the entire trading process — from raw market data to executed orders — with AI embedded at the architectural level.
The Complete Trading Pipeline
Five specialized services work as one system. Each service does one thing with precision, and together they form an autonomous, observable, and auditable trading operation:
| Component | Role |
|---|---|
| Provider | Market data gateway — ingests, normalizes, and serves all market data. Single source of truth. |
| Detector | Strategy runtime — evaluates rules against live data, scores signals with numeric conviction. |
| Advisor | AI decision engine — 8-stage pipeline: market quality gate, ML scoring, conviction calibration, LLM synthesis, spread and R/R validation, execution intent. |
| Inspector | Risk governor — validates every decision against configured policies before any order reaches the exchange. |
| Studio | Operations terminal — real-time visualization of the entire pipeline: signals, AI decisions, risk state, capital efficiency. |
Why AI-Native
Most trading systems add AI as an optional layer. In Barfinex, AI is in the architecture from the start.
The Advisor service runs a structured 8-stage reasoning pipeline on every signal. It checks market quality, applies ML-based conviction scoring, calibrates confidence per market regime, synthesizes context with an LLM, and validates the resulting decision against spread and risk/reward criteria — before passing anything to Inspector.
The Provider exposes its entire API as Model Context Protocol tools. Any LLM can interrogate the full market state, query positions, and inspect running strategies directly.
Every AI decision is telemetered: conviction scores, attribution, regime rotation, hallucination detection, model switching — all published to the event bus and stored in the time-series audit log.
What's Under the Hood
- Data integrity — Provider tracks every candle's lifecycle: gap detection, automated repair, seam validation, ingestion health per connector. 25+ dedicated debug and audit endpoints.
- Observable decisions — Every signal, decision, and risk check is a typed event on the bus. Studio surfaces the full audit trail in real time.
- Risk governance — Inspector manages stop orders, enforces position limits, tracks drawdown, and can throttle or halt signal generation when policies are breached.
- Capital efficiency tracking — Utilization, suppression, and reservation analytics across running strategies.
- Typed event contracts — All channels and payload shapes are defined in
libs/types. No runtime surprises.
How to Start
- Architecture overview — How the five services connect and why.
- Install Provider — The market data gateway; everything else depends on it.
- Install Detector — Define your first strategy as a typed rule configuration.
- Studio terminal — Connect to your running stack and observe everything.