Stock-helper

Moka-chan, a team of LLM agents that writes daily market briefs and sends real-time alerts for US equities.

Stock-helper
TL;DR β€” Meet Moka-chan, a team of LLM agents that reads the market news so I don't have to. She sends a brief before the open, a recap after the close, and a review every Friday, and pings me when something on my watchlist moves.
3scheduled briefs: morning, close, weekly
8layer CIO decision pipeline
30+industries tracked
δΈ­ / ENbilingual reports

The agent pipeline

Every brief runs through a LangGraph pipeline of specialized agents:

🌍 Macroβ†’ 🏭 Sectorsβ†’ πŸ“Š Stocksβ†’ πŸ” Agent picksβ†’ ⚠️ Risk

The final brief is assembled in code in a fixed order, so it is never truncated by the LLM and always has the same structure.

A trading day with Moka-chan

πŸŒ…

7:00 ET Β· Morning brief

Pre-market snapshot, macro backdrop, sectors and stocks to watch, watchlist earnings, and star IPOs.

πŸ””

All day Β· Real-time alerts

Telegram pings for SEC filings, material headlines, and watchlist price moves.

πŸŒ†

16:45 ET Β· Close brief

Closing recap with a Morning vs. Reality section comparing the morning outlook with what actually happened.

πŸ“…

Friday Β· Weekly review

Macro and tracking trends, sector rotation, and next week's earnings.

Thinking long term

Beyond daily briefs, a Market Reasoning Agent writes monthly and biweekly reports that track hypotheses (hold, weaken, invalidated) with evidence scores. A CIO Investment Agent then turns the analysis into an allocation view through an eight-layer pipeline:

Regime→ Theme→ Industry→ Stock→ Portfolio→ Scenario→ Triggers→ Monitor

Each layer follows an evidence β†’ hypothesis β†’ counter-argument β†’ decision pattern.

Roadmap

Today Moka-chan is great at collecting and organizing information. Next, she learns to think and to prove it.

Coming next

🧠 Analysis & strategy model

Move from gathering information to genuine analysis, and from analysis to investment advice.

  • Real reasoning over macro, fundamentals, and sentiment instead of summarizing headlines.
  • Concrete recommendations: what to buy, sell, or hold, and under which conditions.
  • Every call backed by its evidence, counter-arguments, and risks.
Coming next

πŸ§ͺ Trading simulator

Let the agents trade on their own analysis in a simulated portfolio, with no real money involved.

  • Agent decisions executed against real market prices in a paper portfolio.
  • Long-term tracking of returns and risk against benchmarks such as the S&P 500.
  • An honest scorecard of how good the agent really is, without touching real accounts.

Under the hood

Data comes from Finnhub (news, quotes, earnings and IPO calendars), SEC EDGAR filings, and FRED macro series. Models are routed by cost: Gemini Flash-Lite classifies news, and Claude Sonnet writes the analysis, with a per-run budget tracker. Briefs go out through Telegram and HTML email, and Moka-chan also chats in Chinese or English with persistent memory.