What is OpenProphet?

By Jake Nesler. Open source. CC BY-NC 4.0.

OpenProphet is a fully autonomous trading harness that runs an AI agent on a heartbeat loop. The agent wakes up on a configurable schedule, assesses market conditions, manages positions, and executes trades — all without human intervention. A web dashboard streams everything in real time.

It runs locally on your machine and supports any large language model — Claude, GPT, Gemini, and 294+ others via OpenCode.

Architecture

Three components:

Go trading backend — Gin HTTP server on port 4534. Order execution, market data, options chains, news aggregation (Google News, MarketWatch), technical analysis (RSI, MACD), and Alpaca brokerage API integration.
Node.js MCP server — Model Context Protocol server (originally called claude_prophet) exposing 45+ trading tools to any MCP-compatible LLM.
Agent harness — Node.js/Express server running the heartbeat loop, managing OpenCode subprocesses, agent configuration, and the SSE-streaming web dashboard on port 3737.

Web Dashboard (port 3737, SSE streaming)
    |
Agent Server (Node.js/Express, heartbeat loop)
    |
    +--- OpenCode CLI (AI subprocess, any model)
    |        |
    |        +--- MCP Server (Node.js, 45+ tools)
    |
    +--- Go Trading Backend (Gin, port 4534)
             |
             +--- Alpaca Markets API (paper / live)

The Heartbeat Loop

Six market phases with configurable intervals:

Pre-market — 15 minutes. Research, set up watchlist.
Market open — 2 minutes. React to opening volatility.
Midday — 10 minutes. Monitor positions, scan for setups.
Market close — 2 minutes. Close expiring positions, EOD management.
After hours — 30 minutes. Review, log activity, plan next day.
Market closed — 1 hour. Minimal monitoring.

Each heartbeat spawns an OpenCode subprocess with the configured model, system prompt, and MCP tools. The agent calls tools to check its account, scan news, analyze setups, and place orders. Results stream to the dashboard.

The agent can modify its own heartbeat via the set_heartbeat tool — speeding up during volatile periods or slowing down when markets are calm.

45+ MCP Trading Tools

Trading: place_options_order, place_buy_order, place_sell_order, place_managed_position (automated stop-loss/take-profit), cancel_order, close_managed_position.

Market data: get_quote, get_latest_bar, get_historical_bars, get_options_chain, get_positions, get_options_positions, get_orders, get_account.

News intelligence: get_quick_market_intelligence (AI-cleaned MarketWatch), get_cleaned_news, search_news, aggregate_and_summarize_news. Google News + MarketWatch feeds cleaned by Gemini. Reduces raw news by 80-90%.

Technical analysis: analyze_stocks (RSI, MACD, momentum via Go backend).

AI memory: find_similar_setups, store_trade_setup, get_trade_stats. Semantic search over past trades using local 384-dimensional embeddings (sqlite-vec).

Self-modification: update_agent_prompt, update_strategy_rules, update_permissions, set_heartbeat. The agent can modify its own behavior at runtime.

Security and Guardrails

Every tool call passes through a permission enforcement layer:

allowLiveTrading — block all order tools when disabled
allowOptions / allowStocks — asset class gates
allow0DTE — parses OCC option symbols to check expiration dates
maxOrderValue — rejects orders exceeding dollar limit
maxDailyLoss — auto-pauses agent when P&L exceeds threshold
blockedTools — reject calls to specific tools
maxToolRoundsPerBeat — limits tool calls per heartbeat

Multi-Account Sandboxes

Each Alpaca account gets an isolated runtime — its own Go backend process, SQLite database, and agent session. Add paper and live accounts from the dashboard. Switching accounts kills the current backend and restarts with new credentials.

Local-First

Everything runs on your machine. Dashboard, orchestration, config, databases, trade history — all local. External calls go to your LLM provider and Alpaca only.

Getting Started

Requirements: Go 1.22+, Node.js 18+, OpenCode CLI, free Alpaca paper trading account.

Clone from GitHub or use the Premium Setup Guide ($49.99 one-time) for automated installation, 8 strategy templates, 8 agent personas, and setup support.