Harden
Runs custom cybersecurity models on devices to judge every coding agent tool call and stop dangerous ones before they execute
The product is a cybersecurity monitor that runs custom models locally on devices to inspect every tool call a coding agent attempts and blocks, redacts, or logs dangerous actions before execution. It is for developers, engineering teams, and IT teams in businesses that run autonomous coding agents such as Claude Code, Codex, Cursor, and others. It is positioned as a separate pre-execution guardrail outside frontier LLM providers and is delivered as locally-running software installed via a script with on-device decision history.
Key features
- Check tool calls before execution
- Block dangerous agent actions
- Redact and make actions safe
- Log and record allowed calls
- Local decision history on device
- Native hooks for coding agents
- MCP proxy fallback for other tools
- Detect supported agents automatically
- Dashboard for protection activity