AI financial analysis for individual investors — US and Hong Kong equities, crypto, English and Chinese. Live at finlyze.ai.
AI chat
Ask once; the answer comes with how it was madeBTC three-week forecast: 15 agents, three models each give a target — the Agent Run panel on the right shows every stepGARCH volatility forecast, drawn by Python written inside the sandbox for this questionBase / optimistic / pessimistic scenarios from 3,000 Monte Carlo pathsTSLA candlesticks with volume: one CodeAct node fetches, draws and concludes — the note on top is the numeric checkForecast card: most likely price, 80% range, probability of a rise, probability of the level you named, historical hit rate
Analyze
AAPL daily candles with moving averages and Bollinger BandsAI agent report: rating, confidence, macro backdrop, six-dimension scorecard, PDF export
Debate room and personas
NVDA round table: a moderator opens, the personas rebut each other over automatic roundsVote: Buffett HOLD, Wood BUY, Burry SELL, Dalio HOLDTwelve built-in investor personas, or write your own in a paragraph
Watch, watchlist, discovery
Watch: alert rules with browser push; every card has an Ask AI buttonWatchlistAI opportunity scan: real indicator screening over 92 names
Admin overview: SLOs, errors, maintenance mode, feature flagsEight model providers; your own keys come first
How it is built
How an answer gets made
Harness — triage and a clarify gate → a coordinator plans a DAG over 70+ agents → one finalize pipeline (numeric check, qualitative check, answer repair, artifact routing) → answer. What cannot be verified is flagged, never invented.
CodeAct — the model writes one complete Python block; AST allow-list → isolated sandbox → chart QA / code review / log review → repair loop.
Underneath — an event ledger (every run replayable), a point-in-time gate (no look-ahead), cost circuit breakers, and skill incubation → evaluation → promotion with admin approval.
