prismor
Prismor is a self-hosted runtime control plane for AI agents that observes, approves, or blocks rogue tool calls before execution.
Ficha actualizada 22 sept 2026, 18:27 · upstream 22 sept 2026, 18:03 · release v1.54.1
Qué es prismor
Descripción editorial construida a partir del repositorio, su documentación y sus señales de mantenimiento.
AI agents execute shell commands, read and write files, access credentials, and call external APIs autonomously. This creates significant security risks: prompt injection attacks can redirect agent behavior mid-task; malicious content in files or web pages can cause unintended destructive actions; credential files may be exfiltrated during debugging tasks; agents lack complete visibility into spawned subagents; privilege escalation occurs when agents modify sudoers, CI pipelines, or file permissions to resolve errors. Standard OS-level and endpoint security tools monitor the kernel and filesystem but cannot provide the contextual awareness needed for AI agent operations.
Prismor addresses these gaps by providing a comprehensive runtime control plane that sits between the agent framework (Claude Code, Codex, Langchain, Cursor, etc.) and its execution environment. It offers policy enforcement at multiple layers: session-level isolation, tool-call validation before any action occurs, network traffic inspection, and MCP server governance. The platform supports both local deployment on a single machine and cloud-hosted VMs for production-grade agent security.
- Prevent prompt injection attacks by blocking malicious content in user-provided files or web pages from being executed as agent commands
- Detect and block credential exfiltration when agents attempt to read sensitive configuration files during debugging tasks
- Implement network isolation policies that prevent rogue tools from accessing external APIs outside approved domains
- Enforce tool-call restrictions by blocking specific MCP servers or functions before they can be invoked
- Monitor agent behavior across multiple frameworks (LangChain, CrewAI, AutoGen) with a single unified policy engine
- Deploy AI agents in cloud environments (Cursor Cloud, OpenHands Cloud) where the agent runs on a machine you never log into
- Prismor provides full observability of agent activity through local dashboards and live telemetry without requiring agent-side instrumentation
- Supports multiple AI frameworks including Claude Code, Codex, Langchain, Cursor, OpenHands, and many others with a unified policy engine
- Offers MCP server governance that scans tool results for injection attacks before they reach the model's context window
- Provides semantic guardrails using LLM-assisted intent detection to catch paraphrased prompt injections that regex rules miss
- Supports both local and cloud-hosted deployments with portable installation scripts for agents running on remote machines
- Enables layered policy enforcement from session-level isolation down to individual tool calls with granular permissions
- Prismor requires manual setup of the runtime environment, making it less suitable as a plug-and-play solution compared to native agent security tools like Guardrails or Anthropic's built-in protections
- The platform is primarily designed for local deployment; cloud-hosted agents require additional configuration and may not fully leverage Prismor's capabilities without proper integration with cloud infrastructure
- Secret detection relies on file-level scanning rather than comprehensive process monitoring, which could miss secrets stored in environment variables or memory
- Network isolation policies are static rules that must be manually configured; they cannot adapt to dynamic network conditions or new attack vectors
- The platform does not provide automated remediation of detected threats — it only detects and reports findings, requiring manual investigation and response
Perfil de infraestructura
Estimación inicial derivada del repositorio y de los servicios detectados.
Despliegue detectado
Qué señales reales de instalación hemos encontrado.
Install Prismor via pip with `pip install prismor` followed by `prismor setup`. For the Skill, curl, and git-clone alternatives, refer to docs/installation.md. To deploy on cloud agents (Cursor Cloud, OpenHands), use `install-hooks --portable` which creates a safe hook config for commit. Enable live telemetry with `--telemetry-enabled` for enterprise dashboards. Always verify your policy engine is running before deploying agent code.
Stack y dependencias
Componentes inferidos a partir de manifests, configuración y documentación.
Puede sustituir a
Relaciones detectadas por el análisis del repositorio y su categoría.
Experiencia real de la comunidad
Ayuda a contrastar la estimación automática con instalaciones reales.
Todavía no hay datos comunitarios para este proyecto.