Security for AI Agents
AI agents leak secrets, run blind commands, and get hijacked by the data they read. We build the defenses — self-hosted, deterministic, open source. Zero cloud.
IntentWall — Security for On-Chain Agents
Autonomous on-chain agents read their environment and sign irreversible transactions. Poison what they read — a token's symbol, decimals, a price, a recipient — and the decision flips. We proved it across 8 agent frameworks (TS/Python/Rust), plus a Solana and a knowledge-graph agent: the same one-line poison, zero attack code. IntentWall is the defense — a deterministic, out-of-band gate that splits the decision from the authorization and checks rules before signing. It can't be persuaded, because it doesn't read language.
Bastion — AI Agent Security Gateway
A self-hosted local proxy between your AI agents and their LLM providers. It scans every prompt and response for data leaks (20+ patterns — AWS keys, GitHub tokens, private keys, credit cards), detects prompt injection in real time with a local ML classifier, blocks dangerous tool calls before they execute (26 rules — rm -rf, curl|bash, force push, .env reads), and logs everything with AES-256-GCM encryption. DLP · injection detection · tool-call blocking · full audit — zero cloud, fully local.
# FREE — TEST YOUR AGENT
Agent Security Scanner
Is your AI agent vulnerable to indirect prompt injection? Find out in 60 seconds. We generate a honeypot URL disguised as SDK docs, your agent reads it, and we show you exactly which attacks it fell for — canary tokens, injected claims, content suppression, command execution. No install required.
# ALSO FROM AION-0
Forge — AI Workflow Platform
A self-hosted web platform for Claude Code: run AI coding tasks in the background, manage projects from one dashboard, define YAML pipelines, and access everything remotely. Uses your existing Claude Code subscription — no API keys.
Accord — Contract-First Multi-Service Orchestration
Coordinate AI agents across microservices using API contracts as the source of truth. A central Hub dispatches work to services run by AI agents, humans, or both. Built on the A2A protocol.