Skip to content

v1.2.0 — Auth, AI Policies, MCP Server, Slack/PagerDuty, 9 Frameworks

Choose a tag to compare

View all tags
@Justin0504 Justin0504 released this 06 Mar 12:08
· 379 commits to main since this release
v1.2.0
31cbcfb

v1.2.0

Gateway Authentication

Management APIs are now protected by an auto-generated API key. On first boot, the gateway prints your key and stores it in SQLite. The Compliance Cockpit fetches it automatically — no configuration
required for local use. For production deployments, set GATEWAY_API_KEY as an environment variable.

SDK ingest endpoints (POST /api/v1/traces, POST /api/v1/check) remain open so agents work without changes.

AI-Powered Policy Editor

Write security policies in plain English. Open the Policies page, click Describe, type something like:

Block any shell command that deletes or overwrites files outside /tmp

AEGIS calls your configured AI provider and generates a ready-to-save JSON Schema policy. Supports Anthropic (claude-haiku) and OpenAI (gpt-4o-mini). Configure your API key in Settings → AI Assistant.

MCP Server

AEGIS now exposes its audit data as MCP tools at ws://localhost:8080/mcp-audit. Add it to your Claude Desktop config:

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

Available tools: query_traces, list_violations, get_agent_stats, list_policies

Ask Claude: "What did my agent do in the last hour?" or "Which policies triggered the most blocks this week?"

Agent Behavior Baseline

Each agent now has a 7-day behavioral profile: top tools called, risk distribution (LOW / MEDIUM / HIGH / CRITICAL), session count, PII exposure rate, and block rate. Visible in the dashboard when you
click any agent.

Slack & PagerDuty Native Alerting

Alert rules now support three destinations:

  • Webhook — existing behavior, any HTTP endpoint
  • Slack — posts a formatted message with rule name, severity, and current value
  • PagerDuty — triggers an incident via the Events v2 API using your integration key

Configure in Settings → Alert Rules.

Audit Export

Traces can now be exported as CSV directly from the Traces page. The existing PDF export is unchanged.

9 Frameworks, Zero Code Changes

pip install agentguard-aegis==1.2.0

Auto-patched: Anthropic · OpenAI · LangChain / LangGraph · CrewAI · Google Gemini · AWS Bedrock · Mistral · LlamaIndex · smolagents

Assets 2
Loading