Product

Trading Agents Lab

An open-source lab for studying how multi-agent LLM systems reason about financial markets.

What it is

Trading Agents Lab is an open-source desktop application that studies how a team of large-language-model agents reasons about a publicly traded security. A team of specialized agents (analyst, researcher, trader, and a risk-management committee) debates a single ticker and produces a structured recommendation that you can read, question, and learn from. The project calls this multi-agent deliberation The Diligence.

It is a research and analysis tool. It does not place trades, does not connect to brokerages for live execution, and is not investment advice. Every step of the debate is visible, every source is named, and every dollar of API spend is tracked. The transcripts are persisted locally on your machine, not on any server RBJ Global operates.

Built for

People who want to study how AI agents reason about markets, not be told what to buy. Independent researchers comparing model behaviors across providers. Students learning multi-agent system design who want a concrete, non-trivial domain to work in. Anyone curious about what a structured AI debate over a ticker actually looks like.

Why it exists

Software that brings LLMs to financial markets often dresses up a single model as an oracle, or hides its reasoning behind a subscription. Trading Agents Lab does neither. The Diligence is fully readable, every agent's contribution to the debate is visible, every source is named, and every dollar of API spend is tracked. The aim is to learn what a multi-agent system thinks, not to delegate decisions to it.

How it is different

  • Reasoning is the product

    Every agent in the debate is visible. The transcript streams into the UI, the final recommendation carries a confidence score, and the chain of reasoning is auditable. The point is to study the thinking, not delegate to it.

  • Educational, not advisory

    The output is a structured deliberation for research and study. The software connects only to market-data and paper-trading endpoints by design. It cannot place real-money orders.

  • Open source, no accounts

    AGPL-3.0 (with upstream Apache-2.0 attribution preserved). Bring your own API keys, run on your own machine. No accounts, no telemetry, no cloud component RBJ Global operates.

What you can do with it

  • Multi-agent debate

    Specialized agents debate a single ticker and produce a structured recommendation, all visible to you.

    • Roles include analyst, researcher, trader, and a risk-management committee, orchestrated via LangGraph
    • The debate (which the project calls "The Diligence") streams into the UI live so you read the reasoning as it forms
    • Final output is a BUY / SELL / HOLD recommendation with a confidence score, full transcript, and inline disclaimer

  • Bring your own provider keys

    Pick the model you want. Pay the provider directly. No marked-up tokens, no proxying.

    • OpenAI, Anthropic, OpenRouter, and Google Gemini supported out of the box
    • Local models served via Ollama or LM Studio for runs you want to keep entirely offline
    • Keys live in your operating-system keychain; never uploaded, never seen by RBJ Global

  • Market data without paid feeds

    Yahoo Finance covers the default case for free. Alpaca optional when you want the higher-quality feed.

    • yfinance is the zero-configuration default for stocks
    • Alpaca Markets endpoints optional for SIP-feed bars and crypto data
    • Crypto tickers (BTC, ETH, and so on) auto-route to the crypto endpoint

  • Cost Guard

    Hard budget caps and live spend tracking, so a long debate cannot quietly run up the bill.

    • Daily, weekly, and monthly USD caps with live spend bars (green, amber, red)
    • Optional sessions-per-day rate cap for runaway-loop protection
    • Override modal has a 3-second anti-tamper countdown, so unlocking a cap is a deliberate act

  • Local history, webhooks for your own workflow

    Every completed debate is persisted to local SQLite. Send it onward to your own tools if you want.

    • Sessions stored at data/sessions.db on your machine; aborted runs are not retained
    • Replay full transcripts from the History tab, copy as Markdown
    • Webhooks forward completed analyses to a destination you configure (dashboard, notification service); default is no destination

Use cases

  • Comparing model reasoning

    Run the same ticker with GPT-4, Claude, and a local model and study how the debates differ. The Diligence is the same shape; the reasoning is not.

  • Multi-agent systems coursework

    A concrete, non-trivial debate setting for studying LangGraph orchestration, agent-role design, and how consensus forms across specialist roles.

  • Hands-on market literacy

    Reading the bull researcher argue against the bear researcher is the lesson, regardless of whether the ticker ever leaves your watchlist.

  • A baseline to fork

    The Diligence is a working multi-agent design you can clone, modify, and re-run against your own variants. The source is open; the upstream attribution is preserved.

Open source posture

Trading Agents Lab is a fork of TauricResearch/TradingAgents. The upstream code remains under the Apache-2.0 license with attribution preserved in NOTICE; the additions made by RBJ Global are licensed under AGPL-3.0. The full source is available on GitHub. The project is provided as-is, with no warranty. Users are responsible for their own research and decisions.

Standalone trading companion for Clawless

Trading Agents Lab is a standalone trading companion for Clawless Computer. The two products run independently. You can use either one without the other. What they share is RBJ Global's posture across the line: local-first, bring-your-own-keys, no telemetry, and a strong preference for putting reasoning and data on the user's side of the wire.

Free, open source, no account.

Clone the repository, plug in your API keys, run an analysis on your own machine. AGPL-3.0 (commercial users should read the license). For educational and research purposes only; not investment advice.

See the full product site, screenshots, and download instructions:

Visit tradingagentslab.ai →

Source on github.com/RBJGlobal/TradingAgentsLab.

More from RBJ Global

Clawdemy

Free AI education, narrated for life on the move.

Read more·clawdemy.org

Clawless Computer

A calm, capable desktop home for AI work.

Read more·clawless.ai

WhisprDesk

Local-first transcription. Your voice never leaves your Mac.

Read more·whisprdesk.com

ClaudeLink

Coordinate a team of AI coding agents across terminals.

Read more·claudelink.ai

iLoveMD

Free Markdown viewer and converter. Runs 100% in your browser.

Read more·iluvmd.com