AgentKits is a platform that provides production-ready AI agent blueprints with built-in guardrails. The primary function of the service is to offer downloadable system prompts and full kit bundles that developers can use to build AI agents that operate safely in real-world environments. The blueprints emphasize governance features such as human-in-the-loop oversight, worst-case Trust Levels, and reviewable action logs. The platform also includes a risk assessment tool that grades an agent’s safety based on a submitted system prompt and provides a concrete list of fixes. The main features include a library of over 60 free blueprints organized across 30 categories, covering tasks such as access request and provisioning, account research, action item extraction from meetings, bug fixing, contract review, customer support, on-call SRE alert correlation, SOC tier-1 alert enrichment, alert tuning, marketing campaign brief generation, and community triage. Each blueprint documents worst-case actions, failure modes, and Trust Levels. Users can search or browse categories, copy the system prompt, or download the full bundle as files. The platform also offers tutorials and guides on building specific agents and understanding common failure patterns. Typical use cases involve developers and teams who need to deploy AI agents in production with clear safety boundaries. The workflow starts by selecting a blueprint that matches the task, reviewing the design and governance details, then copying the prompt or downloading the kit. The agent can be implemented on platforms such as Claude, OpenAI, LangGraph, or n8n without vendor lock-in. The risk assessment tool allows users to paste a system prompt and receive a governance score and fix list before deployment. The platform also features a learning component where users can earn neurons and build streaks by engaging with content. Technical details include that the blueprints are designed to be model-agnostic and can be adapted to various AI frameworks. The guardrails are encoded as AgentAz Trust Levels, which define boundaries that exist outside the model itself, ensuring that even if the model is jailbroken, confused, or wrong, the boundary holds. The platform does not require a PhD to use, and the tutorials are updated regularly. All blueprints are free to access and download, with no subscription or payment required for the core library.