🤖 Vartovii AI Agents
Vartovii uses a multi-agent architecture powered by Google ADK (Agent Development Kit). The system consists of a root orchestrator and three specialized sub-agents.
Architecture
root_agent (vartovii_orchestrator)
├── corporate_agent — Employer analytics (Trust Score, reviews, vacancies)
├── crypto_agent — Crypto intelligence + on-chain forensics
└── osint_agent — Real-time web research via Google Search
Agents use the configured Gemini model profile and communicate via ADK Function Calling. Production currently uses the Gemini 3.x preview profile with Gemini 2.5 fallback; the stable profile remains available as a rollback path. The orchestrator always delegates — it never responds directly.
1. Orchestrator (Root Agent)
Model: Configured Gemini agent model File: backend/ai/adk_agent.py
The orchestrator routes every user query to the correct specialist. It has no
tools of its own — it uses sub_agents delegation.
Routing logic: | Query type | Agent | |------------|-------| | Company
reviews, Trust Score, vacancies | corporate_agent | | Crypto project, wallet,
smart contract | crypto_agent | | Founder background, news, anything not in DB
| osint_agent |
2. Corporate Agent
Model: Configured Gemini agent model Tools: 6 function tools
| Tool | Description |
|---|---|
search_company | Find company in Vartovii DB (fuzzy match) |
get_trust_score | Trust Score (0-100) with 6-factor breakdown |
list_companies | Top companies sorted by Trust Score / reviews |
compare_companies | Side-by-side comparison of two employers |
get_company_reviews | Recent employee reviews with sentiment |
get_vacancy_intelligence | Ghost job detection, hiring velocity |
Trust Score formula (6 factors):
- Rating Quality (25%) + Sentiment (25%) + Volume (15%) + Source Diversity (15%) + Consistency (10%) + Recency (10%)
3. Crypto Agent
Model: Configured Gemini agent model Tools: 6 function tools
| Tool | Description |
|---|---|
search_crypto_projects | Find project in Vartovii DB |
get_crypto_trust_score | Full Trust Score with tokenomics, investors, TVL |
check_wallet | ETH wallet balance (Ethereum mainnet) |
get_transaction_history | Last N transactions for a wallet |
get_token_holders | Token holder distribution (concentration risk) |
get_contract_info | Smart contract verification status |
Current limitation: Ethereum mainnet only.
4. OSINT Agent
Model: Configured Gemini agent model Tools: GoogleSearchTool
⚠️ Sprint 9 Migration: OSINT Agent is being migrated from ADK
GoogleSearchToolto Vertex AI Agent Builder with native Google Search Grounding. This activates CHF 806 GenAI credits and improves citation quality.
Use cases:
- Founder / CEO background checks
- Recent company news and controversies
- Crypto projects not yet in Vartovii database
- Domain verification, social media analysis
- Any query requiring real-time internet search
Response format: Summary → Bullet details → Recommendation + Source URLs
Related Docs
- Forensic Agent — on-chain BigQuery analysis
- AI Agent Tools API — REST endpoints used by agents