Skip to content

Secure autonomous agents by default.

A modular security boundary for AI coding agents. Run Claude Code and OpenAI Codex in isolated containers — each project gets its own filesystem, credentials, and network. A rogue agent can trash its container but can never escape to the host.

As a user

Download a binary and go. Web dashboard (warden-desktop) opens a browser UI. Terminal UI (warden-tui) runs natively in your terminal. No Docker knowledge needed.

As a developer

Warden’s engine is a Go library and HTTP API. Container lifecycle, worktree orchestration, agent status, network controls, and an event bus — all behind clean interfaces. Build your own UI, CLI, or orchestration layer on top.

Container isolation

Each project gets its own filesystem, env vars, and credentials. No credential bleed, no cross-project file access. Containers are hardened with dropped capabilities, a seccomp syscall filter, and no-new-privileges. Run agents in fully autonomous mode without risking your host.

Network access controls

Per-container network policy: full access, domain allowlist (restricted), or air-gapped (none). Enforced via iptables before any user code runs. Language runtimes (Python, Go, Rust, Ruby, Lua) are auto-detected and open only the required package registry domains.

Real-time agent status

Idle, working, needs permission, needs input, needs answer — across every agent at a glance. Per-project cost tracking with configurable budget enforcement.

Session persistence

Terminals survive disconnects via tmux. Close the tab, agent keeps working. Reconnect later. Attention notifications tell you which agent needs you.

Parallel agents via worktrees

Run multiple agents on the same repo simultaneously, each in its own git worktree. Isolated branches, no merge conflicts, no stepping on each other. Warden orchestrates creation, cleanup, and status across all of them.

Audit and compliance

Unified event logging with activity timeline, summary dashboard, and category filtering. Configurable detail levels (standard/detailed) for cost-volume trade-off. Export audit data as CSV/JSON for compliance review.