Introduction
JARVIS uses a sophisticated multi-agent swarm architecture to gather comprehensive intelligence about individuals in real-time. The system employs specialized agents that run in parallel, each focused on specific platforms or data sources.Architecture
The agent system consists of two main components:1. ResearchOrchestrator (Two-Phase)
A traditional orchestrator that runs static agents in parallel:backend/agents/orchestrator.py
Phase 2: Dynamic URL scrapers for high-value URLs discovered by Exa/Google
2. DeepResearcher (Four-Phase)
An advanced pipeline that streams results as they complete:backend/agents/deep_researcher.py
Runs Exa multi-query search immediately (yields results in ~1s), starts SixtyFour enrichment in background.
Parallel Execution
Orchestrator Approach
backend/agents/orchestrator.py
DeepResearcher Streaming
backend/agents/deep_researcher.py
Agent Base Class
All specialized agents inherit fromBaseBrowserAgent:
backend/agents/browser_agent.py
Key Features
Shared Inbox Pool
Pre-warmed AgentMail disposable email addresses eliminate API latency for login-wall bypass:backend/agents/orchestrator.py
Concurrent Session Management
DeepResearcher uses semaphores to limit concurrent Browser Use sessions:backend/agents/deep_researcher.py
Domain Coverage
Domains handled by dedicated agents (no dynamic scraping needed):backend/agents/orchestrator.py
Data Models
AgentResult
backend/agents/models.py
SocialProfile
backend/agents/models.py
Usage Example
Using ResearchOrchestrator
Using DeepResearcher
Performance
Orchestrator Timings
- Phase 1: 15-30s (6-7 static agents + Exa)
- Phase 2: 10-20s (up to 3 dynamic scrapers)
- Total: 25-50s typical
DeepResearcher Timings
- Phase 0: ~1s (Exa fast pass, immediate results)
- Phase 1: 20-35s (15+ skills in parallel)
- Phase 2: 30-60s (deep extraction + wow tasks)
- Phase 3: 30-90s (verification with account creation)
- Total: 80-180s comprehensive
Next Steps
Browser Use Integration
Learn about Browser Use SDK and Cloud API integration
LinkedIn Agent
LinkedIn-specific agent implementation
Twitter Agent
Twitter/X scraping with twscrape
Deep Researcher
Advanced four-phase research pipeline