OpenAFW is a local agent firewall — keep your secrets off the model, the API relay, and the supply chain. Local credential masking, per-route model routing, and security detectors on the wire.

SOCKS5 and DNS proxy for greywall

Tool-neutral attack corpus for AI agent egress security

What can your AI agent do to the real world? Scan your code. See which tool calls have zero checks

Migrated → github.com/zistica/korveo (rebranded). Archived.

What can your TypeScript AI agent do to the real world? Scan your code. See which tool calls have zero checks

Python verifier for Pipelock action receipts (Ed25519-signed, chain-linked)

Personal AI agent firewall skill for blocking prompt injection, unsafe tool calls, data exfiltration, malicious skills, and memory poisoning.

Policy-based firewall and honeypot middleware for AI agents accessing websites

The identity-native control plane for agent↔tool (MCP) traffic. Every MCP server runs on a private WireGuard mesh — zero open ports — behind an agent firewall: tamper-evident signed audit, policy learned from behavior, identity-gated secrets, and audited cross-org federation. One static Go binary.

GitHub Action for running agent-driven CI jobs through Pipelock with verifiable egress evidence.

Clampd - Runtime Security for AI Agents. Dev infrastructure and orchestration.

Provide an open test corpus to assess and improve AI agent egress security through validated cases and automated workflows.