Agaroham

Context for LLMs

Agaroham (GSV-D1, codename MAHORAGA) is an autonomous, LLM-powered trading agent in GSV's autonomous-agents domain, forked from ygwyg/MAHORAGA and rebranded. It runs 24/7 on Cloudflare Workers, treating a live trading account as a control loop: gather signals, reason over them with a language model, validate the resulting intent against policy, and execute. The loop persists as a Cloudflare Durable Object, so the agent retains state, alarms, and position context across restarts without a dedicated machine or babysitting. The signal layer pulls social sentiment from StockTwits and four Reddit subreddits, with Twitter used as a confirmation source; these are weighted by source and passed to OpenAI, which acts as the analyst, scoring conviction and proposing trades. No trade reaches the broker directly -- a separate policy module gates every decision against a minimum sentiment score, a minimum LLM confidence threshold, position-count and per-position dollar limits, a 2% daily loss limit, and cash-only, long-only constraints. Execution clears through Alpaca, with paper trading as the default mode, making the project explicitly educational rather than a live financial product. The key architectural decision is separating tunable logic from structural logic: trading parameters live in a single harness file annotated with [TUNE], [TOGGLE], and [CUSTOMIZABLE] markers, so adjusting thresholds or adding a new gather*() data source is a local edit rather than a rewrite. Operational surface is an authenticated HTTP API for status and control. Status: the project is in-flight as of early February 2026, positioned in GSV's intercognition-jazz and infrastructure clusters, and runs live on Cloudflare Workers today in paper-trading mode -- it is explicitly framed as educational, not financial advice.