About Knostic
Knostic discovers and secures agents, coding assistants, and their supply chain, from MCP to skills and Code to Cowork. Knostic enforces rules by checking every action against policy before it runs, so allowed actions proceed and risky ones are blocked or sent to a human. Knostic applies least privilege to what agents do, watches every action as it happens, and blocks the unsafe ones before they execute.[1][6]
Kirin is the Knostic product that runs on the endpoint inside supported agents. Kirin sees the tools, extensions, servers, and skills actually installed and used. In one example, Kirin denies a destructive command before it executes, tells the developer why, and logs a Prevented alert for the SOC. AgentMesh rates every skill, server, and extension. Every paid AgentMesh plan includes Kirin IDE security.[4][6][11]
Knostic discovers every agent, model, and AI tool running across endpoints, attributes each to people and teams, and brings them under one policy. Knostic finds agents, extensions, MCP servers, and skills where they run, on the endpoint, rather than inferring from network logs. Discovery is passive, and policy can run in Monitor mode before anything is enforced. Discovered components are rated using AgentMesh verdicts and policy posture. Knostic maps which teams and individuals use which tools and what they can reach.[4][6]
Knostic describes one control plane for all agents, models, and AI tools. Knostic unifies and centrally governs agentic security policies and configurations. Teams can scope what agents may read, write, run, connect to, and install, per team. Knostic describes its guardrails for SaaS, in-house, coding, and Cowork agents.[1][6]
For prompt injection, Kirin catches the injection and blocks the action it would have caused. Kirin inspects what an agent is about to do, not just what it was told. Knostic scans prompts, retrieved context, and tool responses for injection and malicious instructions. Knostic catches credentials in prompts and outputs and blocks or redacts them, in Monitor or Enforce. Teams choose block, downgrade, route to a human, or log per rule and per team. For code, Knostic states that agents model threats before they code, follow secure coding rules, and have every package checked.[1][11]
Knostic keeps an up-to-date list of allowed MCP servers, with every connection validated in real time. Knostic flags deviations from policy or misconfigurations as they occur. AgentMesh continuously rescans the ecosystem, and Knostic research feeds new detections. Decision logs and runtime monitoring provide evidence for audits, incident response, and policy enforcement.[6]
Kirin supports Claude Code CLI, Claude Desktop and Cowork, Cursor, Windsurf, and VS Code + GitHub Copilot, on macOS and Windows, and Linux is in beta. Managers provision the console, push the endpoint client to the endpoints, directly or via an existing MDM, and start in Monitor mode. For Cursor and VS Code, Kirin deploys as an extension and requires the client for shadow AI visibility.
Kirin has three tiers, from individual developers to enterprise-wide rollouts. Knostic provides a package with SOC 2, pen test reports, and a data flow diagram in its trust center. Knostic Inc. is a company incorporated under the laws of the State of Delaware.[4][8][9]
Knostic on the record