Automation
How the autonomous trading agent works
A human-in-the-loop automation stack that fuses market perception, AI reasoning, and cautious execution.
Core Architecture Layers
Perception
Continuous data ingestion from markets, blockchain, news, and internal analytics layers.
Intelligence
Signal fusion, anomaly detection, and scenario analysis assembled into trade hypotheses.
Execution
Paper trading first, then guarded live execution with circuit breakers and queue-backed jobs.
Data Flow at a Glance
- 1
Ingest & Normalize
Goldsky, on-chain fills, and Polymarket APIs stream into DuckDB and Redis-backed pipelines.
- 2
Score Signals
Momentum, whale alerts, Insider Finder, and narrative AI models emit ranked opportunities.
- 3
Approve Decisions
Risk manager enforces position sizing, exposure caps, and manual override checkpoints.
- 4
Execute & Monitor
BullMQ tasks send orders to the CLOB API while telemetry tracks fills, PnL, and health.
Safety Guardrails
- Trading loop ships with a global kill-switch and configurable circuit breakers for volatility spikes.
- All strategies start in paper-trade mode with parity logging before any real capital is allowed.
- Risk config caps wallet exposure, daily loss limits, and requires human acknowledgment for overrides.
- Telemetry sends alerts to Slack/PostHog when jobs stall, balances drift, or model confidence degrades.
Human oversight remains mandatory
Even in live mode, the agent only recommends and stages trades. Operators review, approve, and can pause execution instantly.