Harden Agentic Integrity Foundation runs local cybersecurity models on macOS and Linux to monitor and stop dangerous coding agent tool calls before execution.
What the product does and how it is positioned
Harden Agentic Integrity Foundation is a local-first security layer created to supervise autonomous coding agents. Instead of depending entirely on sandboxes or remote guardrails, it executes specialized cybersecurity models directly on developer workstations to inspect tool calls before they run.
The platform supports environments like Claude Code, Cursor, Codex, and others through native hooks or an MCP proxy fallback. By evaluating commands in real time, it blocks unauthorized actions, suggests safe alternative retries, and records complete session histories locally on disk.
Source-supported ways to use the product
Intercepts destructive terminal or cloud commands, such as deleting critical namespaces, and redirects the agent to safe staging alternatives.
Scans agent tool actions locally to catch and block or redact unauthorized secret exposure prior to tool execution.
Tracks all allowed, modified, and stopped tool calls across coding sessions in a local-only audit store.
The documented workflow, where available
Download and run the local installation script using the provided curl command on a supported machine.
Execute the configuration command to detect supported coding agents present on the machine and activate hooks or proxy routes.
Run connected coding agents normally while the background monitor inspects each tool call, applying allow, block, redact, or retry steering decisions.
Harden Agentic Integrity Foundation operates directly on the host machine to monitor autonomous agent behavior. Rather than relying on cloud proxies or post-execution log parsing, it inserts inspection points directly ahead of tool invocation via native agent hooks or an MCP proxy fallback.
Decisions are determined locally on the machine. Tool calls can proceed unaltered, be logged passively, have sensitive elements redacted, or be blocked outright with recommended alternative commands provided back to the agent session.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
Harden supports macOS and Linux x86_64 for its CLI and daemon, with Apple Silicon and Metal required for the full local model. Windows is currently not supported.
No account or credit card is required to install and use the core local pre-execution features on your machine.
Documented validated agents include Claude Code, Codex, Antigravity CLI, Cursor, Kiro, Hermes, and OpenClaw.
Harden can block the action, redact sensitive components, or supply a safe retry command suggestion back to the agent session.
Audit records, session data, and decision histories are stored locally on the user device without an enforced retention limit.
