Dual-Mode Analysis
SkopaqTrader supports two distinct analysis modes. Understanding when to use each is key to getting the best results.
The Two Modes
1. API Mode (Multi-LLM Pipeline)
The traditional pipeline calls multiple LLM providers to run a LangGraph multi-agent system.
MCP tool: analyze_stock("RELIANCE")
│
┌───────────┼───────────┐
▼ ▼ ▼
Gemini 3.8 Grok 4.6 Claude Opus
(analysts) (social) (risk mgr)
│ │ │
└───────────┼───────────┘
▼
Trade Signal
Key characteristics:
| Property | Value |
|---|---|
| MCP tool | analyze_stock |
| Duration | 2-5 minutes |
| LLM calls | 12-15 separate API calls |
| Providers | Gemini, Grok, Claude, Perplexity |
| Cost | Higher (multiple provider fees) |
| Quality | Each agent uses its specialized model |
When to use:
- Production trading decisions where cost is secondary
- When you want genuine multi-perspective analysis from different model architectures
- Autonomous daemon mode (no human in the loop)
2. Claude-Native Mode (Slash Commands)
Claude itself plays all 15 agent roles, using its own reasoning over raw data fetched via MCP tools.
Slash command: /analyze RELIANCE
│
▼
gather_all_analysis_data (MCP)
│
▼
Claude plays 15 roles:
├── Market Analyst
├── Social Analyst
├── News Analyst
├── Fundamentals Analyst
├── Bull Researcher
├── Bear Researcher
├── Research Manager
├── Trader
├── Aggressive Risk Analyst
├── Neutral Risk Analyst
├── Conservative Risk Analyst
└── Risk Manager
│
▼
Trade Signal
Key characteristics:
| Property | Value |
|---|---|
| Slash command | /analyze |
| Duration | 1-3 minutes |
| LLM calls | 1 (Claude does everything) |
| Providers | Claude only (data from MCP) |
| Cost | Lower (single Claude call) |
| Quality | Consistent reasoning across all roles |
When to use:
- Interactive analysis during a Claude Code session
- Quick iteration on multiple stocks
- When you want to follow Claude's reasoning step by step
- Learning and education (you see each phase unfold)
Comparison
| Feature | API Mode | Claude-Native Mode |
|---|---|---|
| Trigger | analyze_stock MCP tool |
/analyze slash command |
| Speed | 2-5 min | 1-3 min |
| Cost | $$$ (multi-provider) | $ (single Claude call) |
| Diversity | Real multi-model opinions | Simulated perspectives |
| Transparency | Black box (returns summary) | Full reasoning visible |
| Data source | Pipeline fetches its own data | MCP gather tools |
| Memory | Uses agent memories | Uses agent memories |
| Works offline | No | Partially (with Ollama data) |
Mixing Modes
You can use both modes in a single session:
- Start with
/scan(Claude-native) to find candidates - Run
/analyze SYMBOL(Claude-native) for quick assessment - Use
analyze_stock(API mode) for the final decision on top picks - Execute with
/trade SYMBOL(Claude-native) for the actual order
Data gathering is shared
Both modes use the same underlying data sources (INDstocks, Kite Connect, yfinance). The difference is only in who reasons over the data -- multiple specialized LLMs or a single Claude instance.
Configuration
The API mode uses the LLM tier configuration in skopaq/llm/model_tier.py:
# Each role has a fallback chain
"market_analyst": [("google", "gemini-3.8-flash"), ("ollama", "auto")]
"social_analyst": [("openrouter", "x-ai/grok-4.6"), ...]
"research_manager": [("anthropic", "claude-opus-5"), ...]
Claude-native mode requires no LLM configuration -- it uses Claude Code's own model. You only need valid broker credentials for data access.