Fahali
The market AI that grades its own calls — misses included
8 followers
The market AI that grades its own calls — misses included
8 followers
Every market AI claims to be right. Fahali proves when it's wrong: market-risk calls with explicit probabilities and deadlines, publicly graded against what the market actually did — permanent replay links, misses kept forever. 18 detection agents across crypto + US equities/ETFs watch what you hold and explain risk in plain language. Humans use the app; AI agents call it over MCP (Claude, ChatGPT, Cursor) or the fahali SDKs. Read-only — observation, not advice.
The fact that you keep misses forever is honestly really cool, most platforms just bury their bad calls. One thing i'd love to see is some kind of portfolio-level risk score that updates as the 18 agents flag stuff, kind of a single number i can glance at instead of reading every individual alert. would make it way easier to decide when something actually needs my attention.
@glah24980 Thank you — the misses-forever design is the part I refuse to compromise on. A portfolio-level risk score exists today: connect a broker read-only (Binance/Alpaca) and the Portfolio tab rolls your book into one status with VaR (labeled the estimate it is), and — as of a fix that shipped this morning, honestly prompted by feedback like yours — it now names which holdings triggered the flag instead of just counting them. On the API/MCP side, fahali_get_portfolio_risk does the same, and fahali_analyze_custom_portfolio takes any book without connecting anything. Would love to hear what your book looks like — real shapes make the rollup better.
would love a digest mode that pings me only when an agent's confidence shifts by more than X points, so the inbox stays useful instead of noisy. that kind of threshold tuning would make daily checks way more manageable.
@osma_salim25329 You've basically described the design philosophy we shipped this very week: alerts now only page when multiple independent risk factors are elevated and the probability is calibrated against our own graded history — single-sensor observations stay quiet instead of paging you. Two more things you might like: phone notifications only fire for critical/high risk on symbols you actually hold (the firehose never reaches your lock screen), and alerts cluster by symbol + thesis so one situation is one card, not twelve. A true digest mode — "only notify me when my risk state changes beyond a threshold" — is a genuinely good next step and it's going on the roadmap with your name on it.
Love that missed calls stay on the permanent record, that's a rare level of accountability for an AI tool. The plain-language risk explanations actually read like something a thoughtful friend would say, not jargon soup.