An AI framework that mirrors a real trading firm.
Concordal gives each AI agent a specialized role, drives structured debate among them, and mirrors the workflow of an actual buy-side desk. We turn a chaotic single-prompt LLM into an organized, collaborative decision process.
Concordal isn't just a trading framework — it's the future of collaborative AI in finance.
Realistic Firm Simulation
Successfully replicates the collaborative intelligence of a professional trading desk.
Multi-faceted Dialectical Reasoning
Bull / bear debate produces more robust, comprehensive decisions than a single prompt.
Efficient Structured Communication
Solves the information-distortion problem that plagues classic multi-agent systems.
Methodology adapted from github.com/TauricResearch/TradingAgents
Seven specialized roles working in a closed-loop decision process
Inspired by the org chart of an actual hedge fund. Each agent has a specific objective, skillset, and toolset.
Fundamentals · Sentiment · News · Technical
Bull vs Bear researchers
Synthesizes inputs into a sized, conditioned trade
Aggressive · Neutral · Conservative debate
Authorizes the transaction with explicit rationale
Stored, ranked, audit-able. We never auto-execute trades.
Four lenses on one ticker
Fundamental Analyst
Reads financial statements, earnings reports, and insider transactions to estimate intrinsic value.
Sentiment Analyst
Processes social-media posts and sentiment scores to gauge market mood and predict investor behavior.
News Analyst
Analyzes news, government announcements, and macro indicators to assess events that may move the market.
Technical Analyst
Computes technical indicators (MACD, RSI), reads price patterns and volume to forecast direction.
Balancing opportunity and discipline
Trader: synthesize, propose
The trader integrates every analyst report and the bull/bear debate into a concrete trade proposal — with timing and sizing.
Risk Team: evaluate, adjust
The risk team holds its own internal debate (aggressive / neutral / conservative) and adjusts the trade to fit pre-set risk parameters.
Prior decisions can inform the next manager review
Completed decisions are persisted with their ticker and decision-time price. When later price history is available, the system can attach a realised forward return and present recent same-ticker lessons to the manager.
Automated enrichment depends on durable persistence, a reachable market-data adapter and a protected scheduler. If any dependency is missing, the trace reports no injected lesson; it does not invent one.
What the agents actually read
Honest list. We use free public sources today; institutional data feeds (Bloomberg, FactSet, Wind) are on the roadmap.
Cost-aware model routing
Each agent runs at the cheapest tier that still produces good output. Trader and Manager get Pro; analysts/risk run on Flash.
Gemini 3.1 Pro PreviewGemini 2.5 FlashEnd-to-end traceable. Every analyst report, debate turn, and risk vote is stored in the JSON trace.