Kastra
Checks prompts, tool calls, and system requests against policy before AI execution
About Kastra
Kastra is an authorization layer for AI systems that checks prompts, tool calls, Shell commands, and API requests before execution and decides according to policy. Kastra Edge locally governs Claude Code, Codex CLI, and Cursor commands, file edits, and API calls on developer devices. The product provides an OpenAI-compatible proxy, policy as code in the repository, signed append-only audit trails, and cloud, self-hosted, and isolated deployment models.
Best for
Best use cases
Check requests before AI execution
Govern commands and file edits in developer tools locally
Integrate policies through a compatible proxy
Export audit traces to a SIEM
Compare before choosing
The official comparison page says most tools monitor after an AI action, while Kastra makes a decision before the action and positions it as independent infrastructure rather than a feature added to a monitoring product.
Features
Pre-execution policy decisions
Checks prompts, tool calls, Shell commands, and API requests.
Kastra Edge local governance
Checks Claude Code, Codex CLI, and Cursor commands, file edits, and API calls locally on developer devices.
Policy as code and proxy integration
Provides an OpenAI-compatible proxy, with policies stored in the code repository.
Audit and deployment
Generates signed append-only traces and supports cloud, self-hosted, and isolated deployments.
Pros
- 0.8 ms p99 execution-decision latency
- Supports cloud, self-hosted, and air-gapped deployments
- Provides an OpenAI-compatible proxy
- Signed append-only audit trails