egress-control

Open-source security for AI agents: kernel-enforced egress control on macOS and Linux, keys only the operator holds, tamper-evident audit. One command protects Claude Code, Cursor, or any MCP harness. No vendor in the path.

Cold storage that queries like hot storage. High-ratio JSONL compressor with block-level random access. Query cold storage in seconds — no restore, no full decompression. 5–13% ratio on JSONL logs, 25–57% smaller than gzip. 1-hour window via DuckDB in under 6s. Free for personal and open-source use.

Give the agent root in the cell, not on your host.

Open-source governed-sandbox control plane for any workload — per-run identity, brokered credentials, layered egress, approvals, and an append-only audit. Coding agents are the flagship use. Apache-2.0, self-hosted.

Local Docker Compose containment lab for safely experimenting with AI agent workloads behind explicit egress, filesystem, and credential controls.

Identity-aware egress relay for multi-tenant Kubernetes: authorizes workloads by namespace, URL path and HTTP method, then forwards through a corporate proxy under per-tenant credentials.

Redact PII in the browser before a prompt ever leaves the machine, then rehydrate the reply locally: 11 entity types, Luhn/SSN/IBAN checksum validation, and a type-checked Egress Guard on the send path. TypeScript. 173 tests, 100% coverage.

A practical guide for building secure sandboxes using gVisor, JWT authentication, and network egress controls to isolate untrusted code.

FinTrack_App - This application is designed to demonstrate and visualize Kubernetes Ingress (Inbound Routing) and Egress (Outbound Network Security via NetworkPolicies) in a local or production Kubernetes cluster.

AI Agent Security Gateway

Four ways an agent writes to the internet from a sandbox that blocks writes, and the one control that stops all four. Docker Compose, 13 seconds.

Process-attributed eBPF egress leak detection for anonymity gateways

A specification for governing where AI-processed data is allowed to flow. Deployment-agnostic.

Tamper proof enforcement for AI agents on K8s

Reference MCP gateway for ContextBoundary. Deterministic egress and action governance, AARM-aligned Core (R1-R6), CI-verified from a clean clone.

Egress Rate Limiter based on Queen queue and streams

Egress control daemon for AI agents: filters DNS and HTTP CONNECT traffic against a shared allow/deny hostname policy.

Reproducible Docker lab for default-deny AI agent egress and offline Playwright testing

Applying data-loss-prevention egress policy to an un-proxyable realtime audio transport by gating the one server-side choke point — ephemeral-key minting — with a fail-closed tracked-transport downgrade.

Policy enforcement proxy for MCP tool calls: detect leaks, block risky actions, redact sensitive payloads, and log every decision.