Financial LLM Terminal
FoldAlpha connects natural-language questions to real financial data retrieval, showing both the chat answer and the supporting data panel. It runs on proprietary market datasets and a purpose-built financial agent runtime.
KR prices
PeriodSince 2010 · about 16 years
Rowsabout 6.98M rows
Daily Korean stock price data
KR financials
PeriodSince 2004 · about 21 years
Rowsabout 3.21M rows
Annual and quarterly financial statements
US financials
PeriodSince 2013 · about 12 years
Rowsabout 4.88M rows
US company financial statements
KR flows
PeriodSince 2026 · about 2 months
Rowsabout 110K rows
Investor-flow data by Korean stock
Example Questions
Core Strength
Proprietary financial data
FoldAlpha continuously builds and maintains Korean price data, Korean financial statements, US financial statements, and investor-flow datasets. Depth and consistency of the underlying data determine analysis quality.
Core Strength
Purpose-built agent runtime
The runtime was designed after studying commercial agent systems such as Claude Code, Codex, and Cursor, then simplified for financial analysis. The open-source runtime achieved 2.4x faster responses and 100% success rate against the previous system in internal tests.
Forward testing launched
Users can define a strategy in natural language, create a virtual portfolio and scheduled jobs, and evaluate the strategy from the actual operating point forward. Trade history, holdings, average daily return, MDD, and benchmark-relative performance are tracked in one view.
Factor strategy backtesting launched
Users can validate factor strategies built from conditions such as PER, PBR, ROE, and operating-profit growth. CAGR, MDD, Sharpe ratio, and quarterly performance are available in the data panel and archived for review.
Scheduled jobs and Telegram alerts launched
Saved analysis conditions can run automatically at scheduled times, with results delivered through Telegram. Portfolio summaries and condition-based screening alerts can be automated.
Agent runtime open-sourced
The financial agent runtime was built after studying commercial coding-agent architectures such as Claude Code, Codex CLI, and Cursor. It reached 2.4x faster responses and 100% success rate in the internal benchmark and is published with a technical report.
Answers grounded in real data
Important questions are answered after retrieving actual financial and market data. The product is not built to generate plausible text without evidence.
Chat and data panel separated
The center pane explains the result, while the right pane shows the supporting data. The interface is built around answer plus evidence.
Built for Korean investor questions
The current product focuses on earnings, flows, valuation, screening, comparison, and news questions in the Korean-market context.
Automated analysis through scheduled jobs
Register analysis conditions once and run them automatically at scheduled times, including morning portfolio summaries and screening alerts.
Telegram notifications
Receive scheduled-job results and analysis-completion alerts through Telegram without keeping the app open.
Factor strategy backtesting
Combine conditions such as PER, PBR, ROE, and operating-profit growth to validate historical returns in the data panel.
Forward testing
Register a backtested strategy as a virtual portfolio, run scheduled buy and sell simulations, and track live forward performance.
Why This Product
Most financial research requires users to inspect news, earnings, prices, investor flows, and valuation data separately, then combine the evidence themselves.
Financial LLM Terminal reduces that work. It accepts natural-language questions, retrieves the required data, separates interpretation from evidence, and renders the evidence in a data view.
A lightweight agent runtime interprets questions, generates guarded SQL, calls tools, and composes answers on top of 16 years of Korean price data, 21 years of financial statements, and investor-flow data.
The goal is not a generic chatbot. It is an analysis interface shaped around the way Korean investors actually ask questions.
Roadmap
Phase 1
CurrentKorea-focused Perplexity Finance
Stabilize the question-to-data-to-answer flow and complete the Korean-market Q&A experience.
Phase 2
CurrentScheduled alerts and factor backtesting
Run saved conditions on a schedule, deliver results through Telegram, and validate factor strategies through backtesting.
Phase 3
CurrentStrategy backtesting and forward testing
Define strategies in natural language, validate them on historical periods through backtests, then track forward performance with virtual portfolios and scheduled jobs.
Phase 4
PlannedValue-chain and narrative analysis
Map supply-chain relationships and structure market narratives such as themes, momentum, and sentiment.
Phase 5
Long termStrategy research operating system
Manage strategy hypotheses, backtests, forward tests, and failure analysis in one place, then track risk and performance changes across strategies over time.
Vision
The product already includes scheduled jobs, Telegram alerts, factor backtesting, and forward testing on top of a Korea-focused Perplexity Finance experience. The roadmap moves toward value-chain analysis and a research operating system for managing strategy hypotheses, evidence, and forward performance.