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.
Kastra
Freemium
Kastra is authorization infrastructure for AI systems that checks prompts, tool calls, shell commands, API requests, and browser actions before execution. Enforce policy, maintain signed audit trails, and govern local and enterprise AI workflows.
What is 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.
What can Kastra do?
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.
Use Cases
“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.
Frequently Asked Questions
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.
Quick Facts
- 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
Kastra Traffic Analysis
Traffic data is for reference only.
- Monthly Visits
- 2.8K
- Global Rank
- #7,346,278
- User Bounce Rate
- 34.2%
- Avg. Visit Duration
- 00:00
- Pages per Visit
- 1.07
- Domain Rating
- 17
Traffic Trends
| Apr | 0 |
|---|---|
| May | 0 |
| Jun | 2782 |
Traffic Sources
- Direct: 0.00%
- Search: 0.00%
- Referral: 0.00%
- Social: 0.00%
- Mail: 0.00%
- Display Ads: 0.00%
Top Regions
- South Africa: 41.1%
- South Korea: 38.8%
- India: 20.1%
Kastra Alternatives
Agentinel
agentinel.habitwala.in
Agentinel is a local security layer for autonomous coding agents that checks install requests before execution and blocks suspicious npm packages with an offline OSV database and heuristics—without cloud calls.
Panguard AI
panguard.ai
Panguard AI is an open-source security platform for AI agents that scans skills and MCP servers before execution and protects agent actions at runtime, with self-hosted controls and signed compliance evidence for developers, AI vendors, and regulated teams.
Postfleet
postfleet.ai
Postfleet is email infrastructure for AI agents that screens inbound messages, sanitizes content, and extracts clean mail into typed JSON before agents read it. Supports hosted and local MCP access, plus REST for non-MCP workflows.
Execlave
www.execlave.com
Execlave is an AI agent governance and enforcement platform that sits between agents and the systems they access, helping teams add policy checks, approvals, tracing, and audit logs to agents in production.
Lunen
lunen.ai
Lunen is a governed control plane for building and running AI agents across existing business systems. Teams can approve agent actions, manage permissions, and maintain a shared audit record in production.
Phinq
www.phinq.co
Phinq is an open-source runtime governance layer for AI agents. It intercepts tool calls, classifies risk, and pauses higher-risk actions for human approval before execution.