Settings

Gateway configuration, alert rules, and SDK setup

License

Paid features (SSO, custom detectors, longer retention, priority support) unlock when you paste a license key issued after purchase on aegistraces.com/pricing .

Free tier

License key

Deployment Mode

Loading…

Gateway

Gateway URL

Server-side default: set GATEWAY_URL on the Cockpit process (not your SDK). SDKs configure their own URL in agentguard.auto(...).

Status

Checking…

WebSocket

ws://localhost:8080/mcp

Traces API

http://localhost:8080/api/v1/traces

Gateway API Key

Protects the management API. Auto-generated on first start. Paste it here to enable authenticated dashboard access.

Find the key in gateway startup logs: 🔑 Dashboard API key: ...

AI Assistant

Used for natural language policy generation. Your key is stored locally and never sent to the gateway.

Provider

API Key

Once saved, use Describe on the Policies page to generate policies from plain English.

Alert Rules

Rules are evaluated against live traces every 30 seconds. Supports Webhook, Slack, and PagerDuty destinations.

No alert rules — add one below

SDK Quick Start

pip install agentguard-aegis

import agentguard
agentguard.auto("http://localhost:8080", agent_id="my-agent")

# Everything below is unchanged — no decorators needed
client = anthropic.Anthropic()
response = client.messages.create(...)

# Or with env vars (zero code changes):
# AGENTGUARD_URL=http://localhost:8080 python your_agent.py

Anthropic✅ auto

OpenAI✅ auto

LangChain✅ auto

CrewAI✅ auto

Gemini✅ auto

Bedrock✅ auto

Mistral✅ auto

LlamaIndex✅ auto

smolagents✅ auto

Claude Desktop Integration (MCP)

AEGIS exposes audit tools (query_traces, list_violations, get_agent_stats, list_policies) as an MCP server. Add to your Claude Desktop config:

{
  "mcpServers": {
    "aegis": {
      "url": "ws://localhost:8080/mcp-audit"
    }
  }
}

Kill Switch

An agent is automatically revoked after 3 policy violations within 1 hour. Manual revocation:

POST http://localhost:8080/api/v1/agents/:id/revoke

Demo Data

Populate the dashboard with sample data or run the full feature showcase.

Showcase runs 11 steps: safe calls, SQL injection, path traversal, prompt injection, PII detection, data exfiltration, blocking mode, multi-model costs. Watch the Overview tab live.