Universal Agent Monitor

A polished MVP for discovering, monitoring, and reviewing multi-provider agents across endpoints, servers, and cloud environments.

It now models agents generically instead of being locked to one vendor.

Supported sample providers in this repo

• OpenAI / Codex
• Anthropic / Claude
• Google / Gemini
• OpenClaw

Dashboard gallery

Full overview

Fleet pulse and provider mix

Resources and heartbeat monitoring

Agent telemetry cards

Hosts and recent jobs

What it does

The monitor separates two concerns that are often mixed together:

• provider intelligence: Codex, Claude, Gemini, OpenClaw, or custom
• runtime operations: discovery, heartbeat, CPU, memory, disk, network, jobs, and cost

That makes it possible to monitor heterogeneous agent fleets through one normalized dashboard and schema.

Monitoring coverage

• confirmed, observed, and suspected discovery states
• provider-neutral fleet inventory
• normalized runtime metadata with provider, agent_family, and runtime_type
• host CPU, memory, disk, and network telemetry
• per-agent process CPU, memory, disk I/O, uptime, and restart count
• heartbeat freshness with healthy, delayed, and stale states
• hot-host ranking for machines under the highest pressure
• recent job timeline with model, duration, provider, and outcome
• dashboard filters for provider, status, and environment
• raw sample DB export through JSON endpoints

Architecture

flowchart LR
    A["Endpoint Collector"] --> B["Discovery + Metrics API"]
    C["Server / VM Scanner"] --> B
    D["Workspace Observations"] --> B
    E["Kubernetes Scanner"] --> B
    B --> F[(SQLite / Postgres)]
    F --> G["Dashboard API"]
    G --> H["Browser Dashboard"]

Loading

Tech stack

• Node.js 24 with built-in node:sqlite
• plain HTML, CSS, and JavaScript dashboard
• SQLite sample DB for seeded demo data
• Node test runner for API and data verification

Project structure

assets/screenshots/   Dashboard screenshots for the repo
public/               Dashboard frontend
scripts/              Seed and preview generation scripts
src/                  SQLite schema, aggregations, and HTTP server
tests/                API and seed verification
data/                 Local runtime artifacts and generated previews

Quick start

cd C:\Users\ManishKL\Downloads\Codex-Agent-Monitor
npm run seed
npm start

Open http://localhost:3000.

Run tests

npm test

API endpoints

• /api/health
• /api/dashboard
• /api/sample-db

Normalized agent schema

Each monitored runtime is modeled as a provider-neutral agent instance with fields such as:

• provider
• agent_family
• runtime_type
• host_id
• status
• discovery_level
• confidence_score
• cpu, memory, disk, network
• tokens_used, cost_usd
• job metadata

Seeded demo data

The sample seed currently includes:

• 3 hosts
• 4 multi-provider agent instances
• 4 discovery events
• 4 heartbeat snapshots
• 4 jobs
• 3 workspace observations
• 3 host metric snapshots
• 4 agent metric snapshots

Example telemetry in this MVP

• peak host CPU: 67.4%
• average host CPU: 56.1%
• average memory usage: 66.5%
• peak disk usage: 72.3%
• total network throughput: 6260 kbps
• heartbeat state mix: 0 healthy, 0 delayed, 4 stale
• provider mix: openai, anthropic, google, openclaw

Suggested next steps

• real collector ingest API for endpoint and server scanners
• alert rules for stale agents and high resource pressure
• sparkline history for heartbeat and resource trends
• host and agent drill-down pages
• geo and topology views
• RBAC and multi-tenant workspace support