ActPlane is an open-source agent execution and policy enforcement harness developed by the Eunomia-bpf organization. It functions as a security and control plane designed to bridge semantic intent and low-level execution boundaries for autonomous AI coding agents. By combining a dedicated domain-specific language (DSL) with system-level tracing and enforcement mechanisms like eBPF, ActPlane monitors and constrains agent actions at runtime. In standard workflows, ActPlane operates beneath or alongside CLI agent harnesses and code generation systems (such as Claude Code or OpenAI Codex pipelines). It ingests human-defined behavioral constraints, converts natural language policies via its DSL into backend rules, and intercepts OS-level actions—including file operations, process spawning, and network activity—to ensure agent behavior strictly conforms to assigned boundaries. Unlike traditional user-space sandboxes or software wrappers that rely solely on static prompts or API-level permissions, ActPlane enforces policies at the operating system layer using eBPF probes and pinned singleton runtime engines. This architecture eliminates prompt injection vulnerabilities from bypassing system rules, closing the semantic and context gap between high-level agent instructions and low-level kernel executions. ActPlane is an open-source research and engineering project written primarily in Rust and C/eBPF. It is freely available on GitHub for Linux environments requiring privileged kernel-level monitoring capabilities.
eBPF-driven runtime enforcement — Uses kernel-level eBPF probes to intercept, evaluate, and constrain process and network behaviors triggered by AI agents. - Policy DSL engine — Translates natural language constraints and high-level behavioral policies into strict, machine-executable backend representations. - Singleton runtime lifecycle — Manages pinned engine singletons for persistent, low-overhead system monitoring across agent sessions. - Network and I/O tracking — Monitors real-time socket connections (including live UDP/TCP ports) and file system interactions executed during agent tool use. - Semantic gap mediation — Enforces system policies dynamically to prevent LLM agents from deviating from designated task scopes or executing unauthorized system calls. - Rust-based core engine — Provides a lightweight, high-performance evaluation engine built with memory safety and minimal execution overhead.
Use cases
Use Case: Sandboxing Autonomous Agent Terminal Execution When deploying AI agents that execute arbitrary shell commands, developers use ActPlane to prevent unintended filesystem modifications or system-level tampering. ActPlane inspects syscalls via eBPF and terminates or restricts disallowed subprocesses.
Use Case: Enforcing Network Boundaries in Agent Workflows Teams running LLM-assisted code generation tools in production or test environments can configure ActPlane policies to block egress network calls to untrusted endpoints. The engine continuously validates live UDP/TCP connections directly at the OS interface.
Use Case: Translating Natural Language Governance into Machine Rules Security teams can write natural language governance rules for code automation tasks, allowing ActPlane's DSL compiler to map those definitions directly to kernel-level behavioral enforcement maps.
Who it is for
ActPlane is built for systems engineers, AI infrastructure teams, and security researchers who need deterministic, kernel-level guardrails around autonomous AI coding agents running on Linux. It is not intended for standard web developers seeking a standalone interactive chat interface or non-Linux deployment environments.
Install guide
Installation Prerequisites: A Linux kernel supporting modern eBPF features (BTF enabled) and a Rust toolchain. Clone the repository and build the engine via Cargo: bash git clone https://github.com/eunomia-bpf/ActPlane.git cd ActPlane cargo build --release
First Run Running eBPF probes requires root privileges or the CAP_BPF / CAP_SYS_ADMIN capability. Start the ActPlane engine daemon with elevated permissions: bash sudo ./target/release/actplane --config ./config/default_policy.yaml
Example Session Attach ActPlane monitoring to an active agent or execution session to evaluate policies against spawned commands: bash # Run ActPlane policy evaluation against a target process PID sudo ./target/release/actplane attach --pid <AGENT_PID> --policy ./policies/sandbox_strict.dsl
hol-guard — hol-guard is a security scanner and agent guardrail harness developed by Hashgraph Online. Built to safeguard agentic runtime environments and…
Google AX — Google AX is an open-source agent harness and execution orchestrator developed by Google. It provides runtime infrastructure and execution sandboxing…
Skillreaper — Skillreaper is an open-source skill lifecycle management and auditing tool for AI agent ecosystems, developed by thousandflowers. It analyzes agent…
Agent Island — Agent Island is an open-source agent orchestration harness created by developer tristan666666 on GitHub. Designed as a framework for running…
Stratless — Stratless is an open-source agent harness and execution orchestrator developed by the Stratless AI team. It provides a modular infrastructure…
NemoClaw — NemoClaw is an open-source agent orchestration harness and execution framework developed by NVIDIA. It provides structured runtimes, modular skill dispatching,…