Faramesh is an AI Agents & Automation tool that provides execution control and governance for AI agents. It enforces policies, manages human approvals, and offers tamper-evident audits for production AI systems, ensuring agents adhere to defined rules.
Faramesh is an open-source execution control plane designed specifically for AI agents, offering robust governance and security in production environments. It allows organizations to define and enforce policies using its custom language, FPL, or natural language, ensuring every agent action is checked before execution. Faramesh provides three verdicts: PERMIT, DENY, or DEFER for human approval, preventing agents from performing unauthorized or risky actions. It integrates seamlessly with 13 popular AI agent frameworks like LangChain and CrewAI, requiring zero code changes to govern existing agents. Faramesh also brokers credentials through various backends and offers kernel-level enforcement on Linux, making it a critical tool for AI engineers, MLOps teams, and security professionals focused on responsible AI deployment and compliance.
Best used for
Ideal for developers and data scientists who need to ensure their AI agents operate within defined boundaries, prevent unauthorized actions, and maintain compliance. Especially valuable for production AI systems where security, auditability, and human oversight are critical to mitigate risks and ensure responsible deployment.
langgraphlangchaincrewaiautogenpydantic-aigoogle-adkllamaindexaws-strands+ 11 more
Pricing & Plans
Open Source · Likely Not Free
Open Source
FAQs
What is Faramesh's policy language?
Faramesh uses FPL (Faramesh Policy Language), which is designed specifically for agents and includes constructs for sessions, budgets, and mandatory denies. You can write policies directly in FPL or use natural language for compilation. YAML is also supported for policy definition.
How does Faramesh integrate with existing AI agent frameworks?
Faramesh integrates by prepending `faramesh run` to your agent's execution command. It automatically detects the framework, patches tool dispatch, strips ambient API keys, and sets up network interception. It supports 13 frameworks out of the box, including LangChain, CrewAI, and LlamaIndex.
What are the verdicts Faramesh provides for agent actions?
Faramesh provides three verdicts for every agent action: PERMIT (action runs), DENY (action is blocked and agent is informed why), and DEFER (action is held for human approval). If Faramesh encounters an error, the action is always blocked, ensuring it never fails open.
Shypd.aiThe largest open index of AI tools
