AI Agents
Bitcoinology uses a multi-agent architecture where specialized agents handle different phases of the search and analysis pipeline. See Agent Architecture for the full technical diagram.
Agent Roster
| Agent | Icon | Model | Role | Status |
|---|---|---|---|---|
| Oracle | 🔮 | GPT-4o-mini | Intent classification + routing | Routing implemented |
| Jackal | 🐺 | — | Data retrieval (vector + structured) | Retrieval implemented |
| Direct Search | ⚡ | GPT-4o-mini | Fast synthesis from retrieved evidence | Live |
| Playbook | 📖 | GPT-4o | 4-lens deep analysis | Live |
| The Gate | 🚪 | — | Citation verification (constitutional) | Enforced in prompts |
Agent Interaction Pattern
The Gate — Non-Negotiable Principle
"No claim leaves the system without a source citation."
Every response passes through the Gate's validation:
- Every factual claim must map to at least one
EvidenceRecord - If evidence is insufficient, the system returns a safe fallback — never guesses
- Max 2 rewrite attempts before falling back
Feature Flags
All agents are controlled by environment variables:
ENABLE_ORACLE_AGENT=true
ENABLE_JACKAL_AGENT=true
ENABLE_PLAYBOOK_AGENT=true
This allows progressive rollout and A/B testing of agent behaviors.