Show HN: Aegis – Open-source pre-execution firewall for AI agents
AEGIS_JB Saturday, March 07, 2026Every agent framework lets the LLM decide which tools to call at machine speed. There's nothing between the decision and execution — no check, no confirmation.
AEGIS intercepts tool calls before they execute: classifies them (SQL, file, shell, network), evaluates against policies, and either allows, blocks, or holds for human approval.
One line of code, zero changes to your agent:
import agentguard
agentguard.auto("http://localhost:8080")
Built-in detection for SQL injection, path traversal, command injection, prompt injection, data exfiltration, and PII leakage. Every trace is Ed25519 signed and SHA-256 hash-chained.
Supports 9 Python frameworks (Anthropic, OpenAI, LangChain, CrewAI, Gemini, Bedrock, Mistral, LlamaIndex, smolagents), plus JS/TS and Go SDKs.
Self-hosted, MIT licensed, Docker Compose one-liner.
https://github.com/Justin0504/Aegis
Summary
Aegis is an open-source, cross-platform application that provides a secure and decentralized infrastructure for managing digital identities and access control. It aims to empower individuals with greater control over their personal data and online interactions.
2
0
Summary
github.com