18-agent autonomous AI trading bot, debating every trade.

What the platform is

Desktop Wallstreet is an autonomous AI trading platform that runs against an Alpaca paper-trading account on a configurable cycle (default 30 minutes). Each cycle, an 18-agent pipeline scans the market, picks one candidate, runs four parallel analysts on it, runs a bull/bear case-building round (with rebuttals), passes the result to a consensus judge, sizes the position via the trade architect, runs it through the risk analyst, and lands at the Portfolio Manager — the only agent allowed to approve a real order. PM-approved trades route through Alpaca with stop-loss + profit-ladder bracket orders attached.

Every agent's output — bull case, bear case, rebuttals, consensus rationale, risk verdict, PM approval text — is persisted as a decision log. You can read the full reasoning chain for any trade from the live feed.

How the pipeline runs

1. Discovery. Intelligence Scanner ranks 20+ candidates across five signal dimensions (convergence, fundamentals, catalyst, technicals, macro). Market Scout picks the top one, applying sector blocks, earnings blackout, and cooldown rules.
2. Analysis. Four agents run in parallel on the chosen ticker — Technical (RSI/SMA/volume), Fundamentals (P/E, debt, runway), News (catalysts, regulatory risk), Sentiment (social, options flow, institutional positioning).
3. Debate. Bull Researcher and Bear Researcher each build the strongest possible case. Then Bull Rebuttal and Bear Rebuttal cross-counter the other side's three strongest points with specific evidence. No single LLM gets to decide alone.
4. Decision. Consensus Judge weighs the full debate (e.g. 60/40) and emits a conviction score. Trade Architect translates the verdict into a sized position with structured stop + profit ladder.
5. Execution. Risk Analyst checks concentration, sector exposure, macro alignment, and the conviction gate. Portfolio Manager approves or rejects. Execution Agent submits to Alpaca with bracket orders attached. Off-hours submits auto-rebuild as extended-hours limit orders.

See the full agent roster on the homepage.

Risk controls

Six guardrails sit between agent enthusiasm and the broker API: a 15% per-position portfolio cap; sector concentration blocks; a 5-day earnings blackout for new buys; a macro-adaptive cash floor (LOW 10% → CRITICAL 30%); hard stop-loss + trailing stop + profit ladder + time stop attached to every entry; and options Greeks (delta, gamma, theta, vega) tracked on options exposure. The Portfolio Manager is the only agent that can authorise a trade — no agent further down the pipeline can override the PM.

Continuous reevaluation

A separate three-agent reeval pipeline ticks every five minutes against held positions: Macro Analyst checks the global regime, Position Sentinel scores each holding 0-1 on per-position catalysts, and the Reeval PM emits HOLD / TRIM / EXIT verdicts with execution timing (immediate, 1h, 4h, end-of-day, next session). Stops are tightened or loosened automatically based on the macro threat level — no polling, no manual intervention.

Which AI models power it

Per-agent model selection across Anthropic Claude (Sonnet, Opus, Haiku), Qwen (Max, Plus, Turbo), OpenAI (GPT-4o, GPT-4 Turbo), Google Gemini, Groq, xAI Grok, Perplexity, DeepSeek, and a local Claude CLI bridge that routes through your Claude Max subscription. You can run each agent on a different model so no single LLM's blind spot dominates the pipeline. Default flagship config runs Qwen Max on the heavy reasoners and Claude Sonnet on the short-loop agents.

Paper trading only

All trading routes through Alpaca's paper-trading sandbox — there is no live-money path exposed in the product. The platform is built for autonomous AI trading research, not production capital deployment. Each user supplies their own Alpaca paper key + secret on first run; credentials are encrypted at rest in Supabase via AES-GCM.