Overview

The Problem

Building AI agents is hard. Debugging them is harder.

• No visibility — LLM calls happen in a black box. When something breaks, you grep through logs hoping to find the cause.
• Cost surprises — Token usage adds up fast. You only find out when the bill arrives.
• Silent failures — Agents crash or hang with no notification. Users discover problems before you do.
• Privacy concerns — Cloud observability tools want your prompts and API keys. That's a non-starter for many use cases.

The Solution

AgentGazer is a local-first governance platform for AI agents.

curl -fsSL https://raw.githubusercontent.com/agentgazer/agentgazer/main/scripts/install.sh | sh
agentgazer start

One command gives you:

• LLM call tracking — Every request logged with tokens, latency, cost, and model
• Real-time dashboard — See what your agents are doing right now
• Cost analysis — Spend breakdown by provider, model, and agent
• Health monitoring — Heartbeat-based status (healthy / degraded / down)
• Alerts — Webhook and email notifications for downtime, errors, and budget overruns

All data stays on your machine. Prompts and API keys never leave your environment.

How It Works

Point your LLM client at the AgentGazer Proxy (localhost:18900). Zero code changes required.

┌─────────────────────────────────────────────────────────────┐
│                     Your Machine                             │
│                                                             │
│  ┌──────────┐    ┌─────────────────┐                        │
│  │ AI Agent │───▶│ AgentGazer Proxy│───▶ LLM Provider       │
│  └──────────┘    │    (:18900)      │     (OpenAI, etc.)     │
│                  └────────┬────────┘                        │
│                           │                                 │
│                           ▼                                 │
│  ┌─────────────────────────────────┐                        │
│  │   Server (:18880) + Dashboard    │                        │
│  │         SQLite DB               │                        │
│  └─────────────────────────────────┘                        │
└─────────────────────────────────────────────────────────────┘

API Key Management

Store your API keys once, use them everywhere:

agentgazer providers set openai sk-xxx
agentgazer providers set anthropic sk-ant-xxx

Keys are encrypted locally (AES-256-GCM) and never leave your machine. When you use the Proxy with path prefix routing (/openai/..., /anthropic/...), keys are automatically injected into requests — no need to configure each agent separately.

// Point to AgentGazer proxy with your agent name
const openai = new OpenAI({
  baseURL: "http://localhost:18900/agents/my-agent/agentgazer",
  apiKey: "dummy",  // Will be replaced by stored key
});

This also centralizes key rotation: update once with providers set, all agents use the new key immediately.

Multiple agents can share the same Proxy and API key while tracking usage separately — just add the x-agent-id header to each request.

Supported Providers

Provider	Auto-detected
OpenAI	✓
Anthropic	✓
Google	✓
Mistral	✓
DeepSeek	✓
Moonshot	✓
Zhipu	✓
MiniMax	✓
Baichuan	✓

Project Structure

~/.agentgazer/
├── config.json     # Auth token, settings
├── data.db         # SQLite database (events, agents, alerts)
└── lib/            # Installed package (curl install only)

AgentGazer is a monorepo:

Package	Description
agentgazer	CLI — starts server, proxy, dashboard
@agentgazer/server	Express API + SQLite
@agentgazer/proxy	Transparent LLM proxy
@agentgazer/shared	Types, pricing tables, provider detection