Four architectural pillars. No external dependencies. No shared inference. No acceptable points of failure.
Technology Architecture
Data & Market Inputs
Multi-source ingestion, normalized before anything downstream sees it
Cognitive Memory Layer (AAS)
Asset Administration Shells for trades, decisions, and outcomes — append-only, time-stamped
Intelligence Layer (Local AI)
Perception only — no autonomous execution
Governance & Execution Layer
Deterministic, rule-based, fail-closed
Broker & Execution Layer
Multi-broker support with reconciliation
All orders, positions and balances reconciled in real time.
Audit Trail
Evidence Store
Security Stack
Technology Principles
Local-First AI
All models run on your infrastructure.
No Black Box
Every decision is explainable and reproducible.
Evidence-Driven
Trust is earned through verifiable outcomes.
Modular by Design
Agents and components are plug-and-play.
Open & Extensible
REST APIs power the dashboard today. Webhooks and a plugin system are on the roadmap.
Deterministic Execution
Strategies run in code, not in probabilistic text.
Tech Stack Highlights
Sovereign by Design
You own your data. You own your edge.
Transparent & Verifiable
Every step is logged. Every result provable.
Built for Professionals
For systematic traders who demand control.
Extensible Ecosystem
Integrate anything. Automate everything.
Technology that serves your decisions. Not the other way around.
Cognitive Memory Architecture
ZenTrader's Cognitive Memory Architecture is not a session cache or a conversation log.
Each analysis cycle produces a structured digital twin of the agent's decision state.
Cross-cycle continuity means that each new analysis is informed by the full history of prior decisions.
Private Intelligence Layer
ZenTrader's inference engine operates entirely within your deployment boundary.
There is no cloud routing path, no external model dependency, and no third-party API surface exposing your research context.
The result is complete inference sovereignty: deterministic outputs and immunity to provider-side model changes.
Local Inference. Total Sovereignty. Zero External Exposure.
Simultaneous execution cycles are structurally prevented at the process boundary.
Every allocation decision passes through a sequential multi-layer preflight before any execution is permitted.
All broker credentials are cryptographically signed at the API boundary using industry-standard algorithms.
Modular Agent Orchestration
ZenTrader's core intelligence loop is composed of three independent, sequentially orchestrated stages.
The first stage acquires, normalizes, and materializes market inputs into a frozen, validated snapshot.
The second stage receives the validated snapshot and executes the full intelligence pipeline.
The third stage receives a finalized intelligence result and handles all persistence.
Typed Contracts. Stage Isolation. Auditable by Design.
Narrative Intelligence Layer
News, geopolitics, and market narratives are converted into directional, decayed scores per asset and time horizon instead of staying free text an operator has to re-read every cycle.
The model classifies each event against a fixed schema and must cite verbatim evidence for every factor it scores. A deterministic engine, not the model, computes the final number.
Every prompt or model change is graded against fixed reference events before it ships. Calibration history is published, not hidden behind a single confident-sounding demo.
Re-scoring never overwrites a row. Realized price reactions are captured for every score above threshold, so predictive value is measured, never assumed.
Perception-Only LLM. Deterministic Scoring. Auditable Calibration.