High performance command guard and policy enforcement for Agents in Zig
FreeFreeFree tier
About High performance command guard and policy enforcement for Agents in Zig
Orca is a safety layer for autonomous AI agents that evaluates and mediates risky actions — shell commands, file operations, network requests, and tool calls — against a configurable policy. It provides graded mediation (hook, wrapper, proxy, OS-enforced) rather than a universal sandbox, allowing developers to grant agents more autonomy while preventing destructive actions like data deletion, secret leakage, or irreversible changes. Written in Zig for high performance, Orca intercepts actions before execution and can allow, deny, require approval, or log them based on policy. It works with multiple agent frameworks (Claude, Codex, Hermes, etc.) and is free and open source under Apache 2.0.
Key Features
Graded mediation: hook, wrapper, proxy, or OS-enforced protection
Policy-based evaluation of shell, file, network, and tool actions
Prevents destructive commands (rm -rf, git push --force, terraform destroy, etc.)
Prevents secret leakage and unauthorized data access
Supports multiple AI agents (Claude, Codex, Hermes, OpenClaw, Opencode)
Simple CLI with init and run commands
Open source under Apache 2.0 license, written in Zig
Pros & Cons
Pros
- Intercepts dangerous actions before they execute, not after
- Customizable policy rules allow fine-grained control
- Works across multiple agent frameworks with a single policy
- Lightweight and fast due to Zig implementation
- Free and open source with no vendor lock-in
- Graded mediation provides flexibility from monitoring to full enforcement
Cons
- Only covers actions that pass through Orca's mediation path, not a universal sandbox
- Requires policy configuration and integration effort
- May not protect against agents bypassing Orca if not properly configured
- Documentation and community support may be limited as an early-stage project
Best For
Securing autonomous AI coding assistants running on local machinesPreventing accidental data deletion or overwrite by agentsEnforcing access control for shell commands and file operations in CI pipelinesProtecting sensitive credentials and environment variables from agent misuseAllowing safe agent autonomy in long-running workflows
FAQ
What is Orca?
Orca is an open-source safety layer for autonomous AI agents. It intercepts and evaluates risky shell commands, file operations, network requests, and tool actions against a user-defined policy. It is written in Zig and distributed under the Apache 2.0 license.
What actions does Orca protect against?
Orca protects against destructive actions such as git push --force, rm -rf, sudo curl | sh, terraform destroy, kubectl delete, DROP TABLE, aws delete-*, gcloud delete, modifying protected configs, reading .env files, and sending data to unknown hosts.
Does Orca work with any AI agent?
Orca is agent-agnostic and can be used with any AI agent that runs shell commands or uses tools. It provides a CLI command to run agents with guardrails, for example: orca run -- claude, orca run -- codex, orca run -- hermes.
How does Orca enforce policies?
Orca uses graded mediation — hook, wrapper, proxy, or OS-enforced — to intercept actions. Depending on the policy, it can allow, deny, require manual approval, or log the action.
Is Orca free to use?
Yes, Orca is free and open source under the Apache 2.0 license. It can be installed via Homebrew.