# Kastra Kastra è un’infrastruttura di autorizzazione per sistemi di IA che verifica prompt, chiamate agli strumenti, comandi shell, richieste API e azioni del browser prima dell’esecuzione. Aiuta i team ad applicare le policy, mantenere audit trail firmati e governare i workflow di IA locali ed enterprise. - Findings24: - Sito web: - Prezzi: Freemium - Categorie: Assistente DevOps AI, Assistente Codice AI, Strumenti Dev AI ## Cos'è Kastra? Kastra is authorization infrastructure for AI systems. Its main job is to decide what an AI is allowed to do before the action runs, rather than recording or analyzing behavior after the fact. The product checks prompts, tool calls, shell commands, API requests, and browser actions against policy in under a millisecond. The same policy layer can be used across local coding agents, autonomous workflows, and enterprise deployments, with deployment options that include cloud, self-hosted, and air-gapped environments. ## Cosa può fare Kastra? ### Runtime authorization at the action boundary Checks prompts, tool calls, shell commands, and API requests before execution, with the site describing sub-millisecond decision latency. ### Typed, versioned policy authoring Lets teams write policies in Kastra Policy Language, a typed DSL that lives in git, is reviewed in pull requests, and can be tested against fixtures. ### Deterministic policy decisions Supports allow, deny, redact, and escalate or approval-based outcomes depending on the policy and surface being evaluated. ### Cross-surface runtime enforcement Runs enforcement on developer laptops, browser agents, backend services, and other AI surfaces using the same policy model. ### Signed audit trail and exports Produces signed, append-only audit traces that can be replayed and exported to SIEM tools such as Datadog, Splunk, and S3. ### Retroactive review with Recon Includes Kastra Recon, which scans historical agent activity, surfaces risky actions already taken, and drafts policies for them. ## Casi d'uso ### Local coding-agent governance Use Kastra on a developer laptop to gate destructive commands, file edits, and local API calls from coding assistants before they run. ### Autonomous workflow control Apply the same policy layer to autonomous agent workflows so each step is checked and sensitive steps can require human approval. ### Browser-agent enforcement Govern browser-based agents by intercepting clicks, navigation, form fills, and downloads before they fire in the DOM. ### Retroactive policy setup Review historical agent activity with Recon to find risky actions already taken and draft policies before turning on enforcement. ### Enterprise governance and audit Run authorization in enterprise environments where teams want signed evidence, audit exports, and deployment models that include self-hosted or air-gapped setups. ## Domande frequenti ### What does Kastra do? Kastra is positioned as authorization infrastructure for AI systems. It sits in the path of prompts, tool calls, shell commands, API requests, browser actions, and other AI-generated actions, then allows, denies, or in some cases requires approval before the action runs. ### Can Kastra run outside a hosted cloud setup? The source describes deployment models including cloud, self-hosted, and air-gapped. The pricing page also says Enterprise can run in self-hosted, BYOC, or air-gapped environments. ### Which AI tools and agents are mentioned as supported? Kastra’s pricing page lists support for Claude Code, Codex CLI, Cursor, OpenClaw, and any OpenAI-compatible runtime, plus custom agents through SDKs. ### Can policies be tested before they block actions? Yes. The pricing page includes Shadow Mode for observing activity before enforcement, and the policy-engine page describes replaying historical actions against policy versions to see what would have changed. ### Does Kastra offer exports or SDKs? The site says decisions and audit records can be streamed to SIEM destinations including Datadog, Splunk, and S3. The policy-engine page also notes first-party SDKs for TypeScript, Python, Go, Rust, Java, and Swift. ## Informazioni rapide - **Category**: Authorization infrastructure for AI systems - **Deployment modes**: Cloud, self-hosted, BYOC, and air-gapped - **Primary surfaces**: Prompts, tool calls, shell commands, API requests, browser actions - **Named tools/agents**: Claude Code, Codex CLI, Cursor, OpenClaw - **Policy language**: Kastra Policy Language (typed DSL) - **Pricing signal**: Free, Pro, Team, and Enterprise tiers are listed