Autonomous Agents

MDU runs autonomous AI agents that monitor system health and gather requirements through conversation.

Agent Registry

15 agents are seeded in the agent_registry table. Two are currently active:

Agent	Type	Description
Requirement Agent	Conversational	CVO Support Bot for gathering requirements
Log Analyzer	Scheduled	System Guardian for monitoring Docker logs

Requirement Agent (CVO Support Bot)

File: /opt/mdu-api/requirement-agent.js

A conversational agent that helps gather product requirements through natural dialogue using Mistral 7B (via Ollama).

Endpoint

POST /api/admin/agents/requirement/chat
Authorization: Bearer <admin-token>
Content-Type: application/json

{
  "message": "We need a way for users to share models on social media",
  "session_id": "optional-session-uuid"
}

Response:

{
  "reply": "That's a great idea! Let me ask some clarifying questions...",
  "session_id": "uuid",
  "requirement": null
}

When the agent has gathered enough information, it returns a structured requirement:

{
  "reply": "Got it! I've created a requirement for social sharing.",
  "session_id": "uuid",
  "requirement": {
    "title": "Social Media Model Sharing",
    "description": "Allow users to share 3D model previews on Twitter/X and Instagram",
    "priority": "medium",
    "category": "feature"
  }
}

Log Analyzer (System Guardian)

File: /opt/mdu-api/log-analyzer.js

Runs every 5 minutes, analyzing Docker container logs for issues:

1. Fetches recent logs from all monitored containers
2. Analyzes log patterns for errors, warnings, and anomalies
3. Classifies severity: critical, high, medium, low
4. Auto-creates requirements for critical/high issues

Process

Every 5 minutes:
  Docker logs (last 5min) → Pattern matching → Severity classification
     |
     v
  Critical/High → Auto-create requirement in pipeline_jobs
     |
     v
  All findings → Store in agent_runs table

Monitored Patterns

• Container crashes / OOM kills
• Repeated error messages
• API timeout spikes
• Database connection failures
• Disk space warnings

Database Tables

agent_registry

Column	Type	Description
id	UUID	Primary key
name	TEXT	Agent name
type	TEXT	Agent type
status	TEXT	active, inactive, error
config	JSONB	Agent configuration
last_run	TIMESTAMPTZ	Last execution time

agent_runs

Column	Type	Description
id	UUID	Primary key
agent_id	UUID	FK to agent_registry
started_at	TIMESTAMPTZ	Run start time
completed_at	TIMESTAMPTZ	Run end time
status	TEXT	success, failed, timeout
output	JSONB	Run results

Admin Management

Agents are managed via the Admin Hub:

Method	Path	Description
GET	/api/admin/agents	List all agents
POST	/api/admin/agents	Register new agent
PUT	/api/admin/agents/:id	Update agent config
DELETE	/api/admin/agents/:id	Delete agent
GET	/api/admin/agents/stream	SSE real-time updates