AIF works through agent-specific integrations. Select the agent you use, then validate the connection after configuration.
| Agent | Coverage model | Documented boundary |
|---|---|---|
| Claude Code | Native hooks; pre-execution blocking | AIF evaluates covered Claude Code tool calls before they run. |
| Claude Cowork | Claude Desktop integration; pre-execution blocking | Use the latest Claude Desktop app; AIF evaluates covered Cowork actions before they run. |
| Codex | Native hooks; pre-execution blocking | AIF evaluates covered Codex tool calls before they run. |
| Cursor | Hooks; pre-execution blocking | Coverage applies to the configured Cursor IDE and CLI hooks. |
| Antigravity CLI | Hooks; pre-execution blocking | AIF validates the configured Antigravity CLI hook wiring. |
| Hermes | Bridge process; block-and-steer | Unsafe covered actions receive a safe next action when AIF stops them. |
| Kiro | Main-agent hooks; subagent MCP calls via proxy | Kiro subagent shell and file operations are a documented exclusion. |
| OpenClaw | Bridge process; block-and-steer | Unsafe covered actions receive a safe next action when AIF stops them. |
Choose based on your workflow
Use Configure an agent to apply an explicit integration command. If the agent is not listed or an integration changes, check the AIF repository and issue tracker before assuming coverage.
For broader security behavior, data handling, and known limits, read the AI Firewall trust documentation.