Kastra

Kastra sits between AI agents and the actions they try to take, prompts, tool calls, shell commands, API requests, and checks each one against a defined policy in under a millisecond before it's allowed to run. The pitch is that most AI safety tooling monitors what already happened; Kastra is meant to decide what's allowed to happen at all, using a typed, versioned policy-as-code language rather than ad hoc rules. A component called Kastra Edge governs coding agents like Claude Code, Cursor, and Codex CLI directly on a developer's laptop, while an OpenClaw integration extends the same policy checks to browser agents performing clicks, form fills, and navigation. Every decision is written to a signed, append-only audit log that can stream to SIEM tools like Datadog, Splunk, or S3, and a "Recon" feature retroactively scans an AI system's history to flag risky past actions and draft policies automatically. Kastra deploys to the cloud, self-hosted, or air-gapped environments, with SDKs for TypeScript, Python, Go, Rust, Java, and Swift. Kastra Labs Inc. holds SOC 2 Type II certification and is pursuing ISO 27001, targeting regulated sectors like finance, healthcare, and the public sector, alongside a free tier for solo developers. As infrastructure aimed at teams already running production AI agents, it isn't relevant to someone just experimenting with a single chatbot.