Multi-agent
Diligence on any ticker.
Trading Agents Lab is a free, open-source desktop research lab. Twelve specialised AI agents — analysts, researchers, a trader, risk seats — independently study a stock, debate from opposing angles, and converge on a recommendation you can read end to end.
macOS · Windows · Linux · BYO LLM key or run locally
The process is the product.
Most AI trading tools give you a single answer with no idea how it got there. We do the opposite.
A Diligence is a structured deliberation between independent AI specialists. Each agent reads the same facts, plays a different role, and writes their view. Researchers argue bull vs bear. A trader proposes a position. Risk seats critique it from three orthogonal angles. A portfolio manager makes the call.
You read the entire transcript. You see the disagreement. You decide whether the reasoning is sound — long before you decide whether to act on it.
- market_analystLast close $211.50, +19% over 24 sessions. Volume cooling vs 60d avg.
- news_analystCatalyst: GTC keynote framing Blackwell shipments ahead of guidance.
- bull_researcherHyperscaler capex still front-loaded; consensus underwrites Q2.
- bear_researcherCustomer concentration is real — top 4 = 40% of DC revenue.
- traderProposing 0.5% sizing into a partial; let earnings re-rate confirm.
- risk_aggressiveAsymmetric — capex cycle lasts longer than current multiple implies.
- risk_conservativePosition size assumes thesis right; halve it pending guidance.
HOLD · 0.5% partial · revisit post-earnings
Built for serious research, not signal-chasing.
Bring your own LLM
OpenAI · Anthropic · OpenRouter · Gemini · Ollama · LM Studio
Use your existing API key with the provider you trust, or run entirely on-device with Ollama. No keys are stored on our servers — there are no servers.
Real market data
yfinance · Alpaca · headlines · sentiment
Quotes, multi-day price action, news, and crowd sentiment (StockTwits + Reddit) feed every agent's reasoning. Equities, ETFs, crypto, and indices all routable.
Spend that you can prove
cost-guard · per-token math · OAuth-aware
A Cost Guard reserves spend before each run, enforces daily and weekly caps, and shows a live $0.0034 → $0.05 tick during the debate. Subscription and local-LLM runs cost zero — and the UI says so.
The full transcript
phase chips · agent-by-agent · copy markdown
Every Diligence is rendered live with phase progress, agent labels, and color-coded reasoning. Save the markdown; replay it later from History.
Outbound webhooks
telegram · slack · discord · generic JSON + HMAC
When a debate finishes, push the decision to Telegram, Slack, Discord, or your own endpoint. Wire that endpoint to your broker on your terms — we never see the trade.
Zero telemetry
no analytics · no accounts · no install ping
The desktop app makes no network calls beyond the providers you configure. The marketing site you're reading right now has no analytics either. We don't know you're here.
Twelve agents. Four phases. One transcript.
The same pipeline every research desk runs informally — codified, paralleled, and made visible.
Analysts
- market_analystprice action · volume
- news_analystrecent headlines
- fundamental_analystmetrics · context
- sentiment_analystsocial signal
Researchers
- bull_researcherthesis for
- bear_researcherthesis against
- research_managerarbitrates
Trader
- traderproposes position
Risk
- risk_aggressiveupside view
- risk_conservativedownside view
- risk_neutralmiddle path
- portfolio_managerfinal call
Analysis, not execution.
Trading Agents Lab will never place a trade on your behalf — not via paper trading, not via a hidden flag, not even if you ask it to. We build the analysis layer; you keep the brokerage relationship.
This isn't a temporary scope choice. It's a deliberate regulatory firewall: when execution code never exists in our repository, there is structurally nothing for the SEC to regulate as a trading platform. You bridge the analysis to your authorised broker via a webhook on your terms.
Read our positioning →