Agent Runner (Proposed)

Caution

The Agent Runner is a proposed new service — not yet deployed. This document describes the planned architecture. See AI Agents: Roadmap for the implementation timeline.

Overview

The Agent Runner is a new Node.js/TypeScript service that enables background agentic AI execution. It polls the Backend API for scheduled agents, registers cron jobs, and fires the AI Service chat API on schedule.

Setting	Value
Technology	Node.js + TypeScript + Express
Port	9400 (main only)
Status	proposed — no container, not deployed on either estate
Repo	Deployd-io/logrite-agent-runner (to be created)

Responsibilities

1. Poll Backend API every 60 seconds for active scheduled agents
2. Register cron jobs for each agent based on its schedule
3. Execute agents by calling AI Service /api/v1/chat/sync with:
   • Agent’s system prompt
   • Allowed tools list
   • Context query
4. Store execution results via Backend API
5. Alert via Discord if execution triggers alert conditions

Planned API

Method	Path	Description
GET	/health	Health check
GET	/api/v1/status	Active cron jobs, last execution times
POST	/api/v1/trigger/:agentId	Manually trigger an agent

Execution Flow

flowchart TD
  A[Startup] --> B[Poll Backend for scheduled agents]
  B --> C{New agents?}
  C -->|Yes| D[Register cron jobs]
  C -->|No| E[Wait 60s]
  E --> B
  D --> F[Cron fires]
  F --> G[GET /agents/:id/context from Backend]
  G --> H[Build system prompt + toolsAllowed]
  H --> I[POST /chat/sync to AI Service]
  I --> J[Receive response]
  J --> K[POST /agents/:id/executions to Backend]
  K --> L{Alert condition met?}
  L -->|Yes| M[POST Discord webhook]
  L -->|No| N[Done]
  M --> N

Integration Points

Service	Interaction
Backend API	Read agent configs, write execution logs
AI Service	Call /api/v1/chat/sync with agent context
Discord	Alert notifications via webhook

Configuration

The Agent Runner reads its config from the Backend API — no separate config file needed. Agent schedules, system prompts, and tool permissions are all stored in the Backend database.

Running it (when built)

# Start via master CLI
./cli/logrite-manager.js start agent-runner main

# It would be a compose service like every other, with no supervisor
docker logs -f agent-runner
# restart / redeploy through https://management.logrite.ai

Next Steps

See:

• AI Agents: Architecture — Full system design
• AI Agents: Execution Loop — Detailed execution flow
• AI Agents: Roadmap — Implementation phases