Current exploration · Trading Agent · v5.1

Zero Effort
Trading Agent.

An autonomous trading system that monitors nine assets, makes LLM-driven decisions, and executes via broker API, with a live dashboard tracking every move. No manual intervention. No hand-holding. Just a system that runs.

v5.1
Current Version
9
Assets Tracked
3
Broker Accounts
24/7
Autonomous
60s
Refresh Cycle
Lifecycle

From a script to a system.

ZETA didn't start as an agent. It started as a question, became a script, grew into a pipeline, and evolved into something that trades on its own. Five iterations, each one teaching the next.

v1.0 The Script
"Can a script trade?"
A single Python file. Hardcoded if-then rules. One asset, one account, one direction. It bought when the RSI was low and sold when it was high. It worked, sometimes. It broke, often. But it proved the concept: automated trading was possible without a Bloomberg terminal.
Python RSI 1 asset if-then
v2.0 The Pipeline
"Can it scale?"
The script became a pipeline. Broker API data flowed through n8n workflows into a Teable database. Positions, P&L, and win-rates were normalized and stored. Three accounts, nine assets. Still rule-based, but now with data to prove which rules worked, and which didn't.
n8n Teable 9 assets 3 accounts
v3.0 The Algorithm
"Can it think?"
The first LLM experiment. Instead of hardcoded rules, a language model analyzed market data and output structured decisions. The results were logged with reasoning, not just "buy" or "sell," but why. The rules were gone. The model was wrong often enough to be humbling, but right often enough to be interesting.
LLM Structured output Decision log
v4.0 The Agent
"Can it run itself?"
Hermes Agent replaced the cron job. The system now had a persistent runtime, error handling, and the ability to recover from failures without human intervention. Win-rate monitoring was added. Directions that exceeded a 10% loss rate were paused automatically. The dashboard went live.
Hermes Agent Auto-recovery Live dashboard Win-rate guardrails
v5.1 The Self-Learner
"Can it learn from itself?"
Closed trades are matched back to their decision records. P&L is attributed per asset. Win-rates roll on a 7-day window. The system sees its own results and adjusts. Not through backtests, through live performance. Two LLM layers: one for trading decisions, one for review and analysis. The loop is closed.
Closed-loop learning Dual LLM OpenRouter Ollama Cloud
Architecture

How it works.

Broker API
n8n Workflow
fetch positions, win-rates, account data
Teable (DB)
Normalize Node
merge, compute 7d win-rates, status
LLM Decision
LLM via OpenRouter, analyze, decide, output
Hermes Agent
Broker API Execution
open / close / hold per asset
LLM Review
Second LLM via Ollama Cloud, review decisions, analyze outcomes
Live Dashboard
real-time positions, P&L, win-rates, decision log
What it does

Features.

Monitoring

Nine assets, continuous tracking

ZETA monitors nine indices and commodities across three broker accounts. Each asset has its own position, direction, margin, and P&L, updated every 60 seconds.

Decisions

LLM-driven, not rule-based

Instead of hardcoded if-then rules, ZETA uses large language models via OpenRouter to analyze market conditions and make trading decisions. Every decision is logged with its reasoning : not just the action, but why.

Review

A second model checks the first

Decisions don't go unchecked. A second LLM layer via Ollama Cloud reviews trading decisions and analyzes outcomes. catching edge cases the primary model might miss. Two perspectives, one loop.

Risk

Built-in guardrails

Each asset can be toggled on or off independently. Win-rate monitoring pauses directions that exceed a 10% loss rate. Margin closeout is tracked in real-time. The system can stop itself, no human needs to pull the plug.

Dashboard

Full transparency

A live dashboard shows everything: positions, unrealized P&L, 7-day win-rates, margin usage, and a decision log. Dark mode, responsive, with win-rate mini-bars and P&L micro-charts. Built as a single HTML file, no framework, no dependencies.

Outcome

Closed-loop learning

Closed trades are matched back to Teable records, P&L is attributed per asset, and win-rates are recalculated on a rolling 7-day window. The system learns from its own results. not from backtests, but from live performance.

Tech stack

What it's built with.

Broker API n8n Hermes Agent OpenRouter Ollama Cloud Teable Python Docker Traefik nginx Vanilla HTML/CSS/JS

ZETA is a system we built for ourselves. It manages real capital on real accounts. We don't sell signals, promise returns, or offer financial advice. This page documents what we built, not what you should do.

Back to L.A.B. Who we are