Back to Projects
agentactive
PRISM AI Agent System
54-module autonomous AI system with 33 agents, sovereign command layer, email triage, market intelligence, and session recap.
The PRISM AI Agent System is a 54-module autonomous platform with 33 agents that runs the operational backbone of PRISM AI Consultants and VersAssist. It operates continuously with minimal human intervention.
Architecture
The system is built in Python with 65 CLI commands, running across a local Mac environment and a Hostinger VPS. It maintains shared state across four agent personas (CEO, CMO, CFO, COO) and processes 880+ indexed transcripts.
Core Components
- Sovereign Agents - strategic command layer that sits above all advisory agents. PRISM Sovereign handles revenue command; VersAssist Sovereign handles delivery command. Both enforce math-based business thresholds.
- Email Triage - runs three times daily on the Hostinger VPS. Haiku classifies incoming emails by priority and category. Sonnet drafts responses. Most routine emails are handled without human review.
- Market Intelligence - scans 12 RSS feeds twice daily and produces briefings on industry trends, competitor activity, and relevant news.
- Session Recap - operates on 30-minute cycles tied to calendar events. Captures meetings, extracts summaries, and logs action items automatically.
- Slack Bot - an agentic chief-of-staff accessible via Slack. Handles quick queries, status checks, and operational requests.
- Watchdog - monitors all running services and alerts when something breaks or goes offline.
Technical Details
- 54 Python modules in the source directory
- 20 automation runners
- Shared state maintained across 21 state files
- Separate OAuth tokens and state directories for PRISM and VersAssist
- Self-hosted on macOS and Hostinger VPS (Docker/n8n)