/ → Dashboard (home)
/strategies → Strategy list + create
/strategies/:id → Strategy detail
/quick-ask → Single/multi ticker analysis
/quick-ask/:conversationId → Saved conversation
/backtest → Backtest configuration + results
/memory-lab → Strategy memory & learning dashboard
/insights → Daily investment suggestions
/approvals → HITL approval panel
/scanner → Market scanner
/watchlist → Watchlist management
/risk → Risk analytics
/reports → Research report generator
/settings → System configuration
Purpose: Landing page. Overview of everything at a glance.
Content:
- Header row: Date, market status (open/closed), quick ticker search bar
- Strategy leaderboard (top 6 cards): Ranked by return %. Each card: strategy name, type badge (agent/quant/hitl), total return %, benchmark comparison (+3.2% vs SPY), mini sparkline, status indicator
- Multi-line performance chart: Equity curves of top 5 strategies overlaid with benchmark line (SPY/CSI300), selectable time range (1w/1m/3m/all)
- Today's market brief: Collapsed card with morning insight summary. Click to expand full text.
- Alert panel (right sidebar or bottom): Recent triggered alerts — price alerts, agent flags, HITL pending count with red badge, memory reflections ready for review
- Quick actions bar: [New Strategy] [Quick Ask] [Run Backtest] [View Approvals] [Memory Lab]
Empty state: If no strategies exist, show onboarding cards: "Create your first strategy", "Try Quick Ask", "Explore the scanner".
Content:
- Filter bar: by type (agent/quant/hitl/all), by status (active/paused/stopped/all), by tag, search by name
- Strategy table/cards: Name, type, status, tickers, return %, excess return vs benchmark, created date, last run time
- Action buttons per strategy: Start/Pause/Stop, Edit, Clone, Delete (with confirmation)
- [+ New Strategy] button → opens strategy creation flow
Strategy Creation Flow (modal or new page):
- Basic info: Name, description, tags
- Type selection: Agent / Quant / HITL (with description of each)
- Ticker selection: Multi-select from predefined groups + manual input
- Trading Beliefs (for Agent/HITL):
- Natural language belief editor: add one or more trading philosophies
- Examples provided: "专注于短期事件驱动", "偏好低估值蓝筹"
- Each belief spawns a Research Agent perspective
- Belief weight configuration (initially equal, auto-adjusted by performance)
- Agent Configuration:
- Agent enable/disable toggles (checkboxes grouped by team):
- Analysts: Market, News, Fundamentals, Sentiment, Technical, Macro
- Researchers: Bull, Bear (debate rounds: 1-5)
- Risk: Aggressive, Safe, Neutral (risk debate rounds: 1-5)
- Custom prompt template (optional override)
- Model selection (quick-think + deep-think), temperature
- Cross-review toggle (2nd LLM reviews high-risk decisions)
- Agent enable/disable toggles (checkboxes grouped by team):
- Quant Configuration (for Quant type):
- Select strategy from dropdown (MA Cross, RSI Reversal, TS Momentum, Vol Targeting, Risk Parity)
- Alpha Zoo factor selection: browse 452 pre-built factors, select relevant ones
- Parameter inputs
- Schedule: Frequency (hourly/daily/weekly), execution time
- Memory & Learning:
- Memory enabled toggle
- Weekly reflection toggle
- Memory recall limit (max past decisions to inject per run)
- Review & Deploy: Summary card → [Deploy Strategy]
Content:
- Header: Strategy name, type badge, status badge, [Pause/Resume] [Edit] [Clone]
- Performance tabs:
- Overview: Equity curve with benchmark overlay (SPY/CSI300), key metrics cards (total return, excess return, Sharpe, max drawdown, win rate, total trades), benchmark comparison panel
- Debate Viewer (new): Visual timeline of agent debates. Each round shows:
- Investment debate: Bull claim vs Bear claim side-by-side
- Risk debate: 3-column layout (Aggressive / Safe / Neutral)
- Token-level streaming replay available
- Click agent card → expand full argument
- Holdings: Current positions table (ticker, quantity, entry price, current price, PnL, weight %)
- Trade History: Filterable table of all trades
- Decisions: Timeline of agent decisions — each entry shows direction, confidence, winning belief, report excerpt. Click to expand full analysis including debate history and cross-review result.
- Memory Lab (new): Strategy-specific memory view:
- Recent memories (OWM-scored)
- Weekly reflections timeline
- Knowledge base (rules / findings / failures)
- Active hypotheses and their status
- Approvals (HITL only): Approval history log with cross-review details
- Audit Trail: Event timeline with expandable nodes
- Settings tab: View/edit strategy configuration (same form as creation but pre-filled)
Empty state (new strategy): "Waiting for first execution..." with next scheduled run time.
Content:
- Left sidebar: Conversation history list (saved + recent). Each item: auto-generated title, date, tags. Search/filter.
- Main area:
- Input bar: Ticker input + optional natural language question + date picker (default: today)
- Analysis mode selector: Fast (market + news only) / Standard (all agents) / Deep (debate mode with cross-review)
- Agent toggle panel (collapsible): Quick enable/disable individual agents for this analysis
- Belief selector (optional): Pick which trading belief to apply for this analysis
- Progress display: During analysis, show agent status cards with streaming progress:
[company_overview] ✅ Complete (1.2s) [market_analyst] ✅ Complete (3.2s) [news_analyst] 🔄 Fetching news... [fundamentals] ⏳ Queued [sentiment] ✅ Complete (2.1s) [technical] ✅ Complete (1.8s) [macro] ⏳ Queued ────────────────────────────── [bull_researcher] 🔄 Debate round 1/2... [bear_researcher] 🔄 Debate round 1/2... ────────────────────────────── [risk_analysts] ⏳ Waiting for upstream - Result display: Decision card (direction, confidence, timeframe, one-line summary) → debate visualization (collapsible: bull vs bear arguments) → expandable full reports per agent → news sources with links
- Bottom actions: [Save to History] [Add Tags] [New Follow-up Question] [Export PDF]
- Multi-ticker comparison: "Compare AAPL vs MSFT" → side-by-side result cards
- Topic analysis: Describe a sector/theme → system auto-discovers relevant tickers → batch analysis
Same layout as Quick Ask but pre-loaded with saved conversation history. Supports follow-up questions that maintain context from previous analyses in the same conversation. Past debate records are preserved and viewable.
Content:
- Configuration panel (left or top):
- Strategy selector (existing + "ad-hoc parameters")
- Ticker(s) selector
- Date range (start/end with calendar picker)
- Forward validation period (7/14/30/60/90 days)
- Benchmark selector: SPY / CSI300 / custom ticker
- [Run Backtest] button
- Results panel:
- Summary metrics: Total predictions, accuracy %, cumulative return vs benchmark, excess return, information ratio, Sharpe ratio
- Timeline chart: Each prediction plotted on price chart — green dot = correct, red dot = wrong, dot size = confidence. Benchmark line overlaid.
- Confusion matrix: Predicted vs Actual direction (Bullish/Bearish/Neutral)
- Per-period breakdown table: Date, ticker, predicted direction + confidence, actual direction + return, was correct?
- Export: CSV of all results, PDF report
- Parameter grid backtest: Run same strategy on multiple tickers or multiple date ranges in parallel
Purpose: Central hub for strategy learning and knowledge accumulation. Inspired by TradeMemory Protocol + QuantGPT knowledge base.
Content:
- Strategy selector: Pick which strategy to view memory for
- Memory Overview tab:
- OWM score distribution chart (histogram of memory quality)
- Recent memories table: ticker, action, outcome, OWM score, date
- Click memory → expand full record (all 5 layers)
- Search memories by ticker, date range, outcome
- Reflections tab:
- Weekly reflection timeline
- Each reflection card: win rate, drawdown, behavioral diagnostics
- Trend chart: win rate over time (strategy health monitor)
- Strategy decay alerts
- Knowledge Base tab (from QuantGPT):
- Rules (verified, must follow)
- Findings (empirical discoveries)
- Failures (falsified paths, avoid)
- Add/edit/delete entries
- Link entries to specific sessions/memories
- Hypotheses tab:
- Active hypotheses with status badges
- Evidence table per hypothesis
- Create new hypothesis → define claim + acceptance criteria + budget
- Resolve hypothesis (confirm/reject/stale)
Content:
- Calendar view: Monthly calendar with dots on days that have insights. Click a day → expand that day's morning/midday briefs.
- Today's briefs: Morning brief card + Midday update card (if available). Full text with key events listed.
- Accuracy tracking: For each past insight, show whether the directional call was correct. Overall accuracy % chart.
- Channel status: Show which notification channels are active (Email, Telegram, WeChat, Feishu, Discord, Slack)
- Email preview: Show what the email looks like (HTML preview)
- Event alerts: Chronological list of event-driven alerts
Content:
- Pending queue: Cards sorted by urgency. Each card:
- Strategy name, ticker(s), decision summary, triggered rules, time remaining
- Cross-review badge: shows whether 2nd LLM agreed or disagreed
- [Approve] [Reject] [Modify]
- Review detail (click a card to expand):
- Full context: all agent reports + debate history for this decision
- Decision details: direction, position change, confidence
- Cross-review panel: 2nd LLM's independent analysis + consensus/divergence indicator
- Triggered rules (highlighted)
- Modify form: Adjust target position %, override direction, add notes
- History tab: All past approvals, filterable
Empty state: "No pending approvals" with green checkmark.
Content:
- Mode toggle: Rule-based / Agent-driven / Belief-driven (new)
- Rule-based mode:
- Condition builder: [+ Add Condition] → field dropdown, operator, value
- Saved queries list
- Results table: ticker, current price, matched conditions, action links
- Agent-driven mode:
- Natural language input
- Agent reasoning display
- Results with explanations
- Belief-driven mode (new):
- Select a trading belief from existing strategies
- System auto-generates scanner query matching that belief's philosophy
- Export: CSV download of results
Content:
- Multiple watchlists (tabs): "My Positions", "AI Stocks", custom...
- Per-watchlist: Table (ticker, price, change %, RSI, MACD signal, news sentiment badge, alerts). Click ticker → Quick Ask popover.
- Add ticker: Search + add to list
- Alert management: Per-ticker alert configuration. Alert type + threshold.
- Drag reorder: Tickers
Content:
- Portfolio selector: Pick a strategy to analyze
- VaR/CVaR: Value at Risk chart (histogram of historical returns with VaR line)
- Stress test:
- Preset scenarios: "2008 Financial Crisis", "2020 COVID Crash", "2022 Rate Hikes"
- Custom scenario: "Market drops X%, VIX spikes to Y"
- Results: estimated portfolio impact
- Correlation heatmap: All holdings pairwise correlations
- Sector concentration: Pie chart of sector allocation
- Drawdown chart: Historical drawdown curve with peak-to-trough annotations
Content:
- Report type selector: Stock Deep Dive / Sector Analysis
- Stock Deep Dive: Ticker input → Generate. Agent runs full analysis + extended research. 10-15 page PDF with charts, tables, analysis text.
- Sector Analysis: Sector/theme description → system discovers tickers → batch analysis → aggregated report
- Report history: Previously generated reports, click to re-download PDF
- Report template: Select sections to include (Technical, News, Fundamentals, Sentiment, Risk, Recommendation, Debate Summary)
Content:
- LLM Configuration:
- API key (masked), base URL, model selection
- Quick-think model + Deep-think model (separate)
- Test connection button
- Agent Defaults:
- Default active agents (checkboxes)
- Default debate rounds
- Cross-review default (on/off)
- Skill Management (new):
- List all skills (builtin + user-created)
- Enable/disable skills
- Create new skill: name, category, prompt template
- Edit existing skill (user-created only, builtin are read-only)
- Skill version history
- MCP Integration (new):
- MCP Server status (running/stopped)
- Transport mode (stdio/SSE/streamable-http)
- External MCP servers list: add/remove/configure
- Tool listing per server
- Data Sources:
- Yahoo Finance status indicator
- Google News status
- AkShare status
- Cache TTL setting
- Notifications (expanded):
- Email: SMTP host, port, credentials, test button
- Telegram: Bot token, chat IDs, test button
- Enterprise WeChat: Webhook URL, test button
- Feishu: Webhook URL, test button
- Discord: Webhook URL, test button
- Slack: Bot token, channel ID, test button
- Memory & Learning:
- Memory retention period (days)
- Weekly reflection schedule (day + time)
- Clear memory (per strategy)
- Scheduler:
- Global execution frequency limits
- News fetch interval
- Data Management:
- Export all data (JSON dump)
- Clear cache
- Delete strategy data
- About:
- Version, team, licenses
- Link to proposal PDF