Phantasm empowers you to add a human-in-the-loop approval layer to your AI workflows. With our open-source toolkits, you can easily monitor and manage AI performance in real-time, ensuring critical decisions are vetted before execution. Ideal for applications where accuracy matters, Phantasm integrates with any AI framework for a reliable automated process.
Phantasm is an innovative open-source toolkit designed to create a human-in-the-loop (HITL) approval layer, allowing you to monitor and guide AI agents' workflows in real-time. With Phantasm, you can seamlessly integrate a human approval process into your AI's operations, ensuring accuracy and accountability while working with any AI framework or model.
Key Components
Phantasm comprises three essential components that collaboratively enhance your HITL experience:
- Server: Coordinates HITL workflows between human approvers and AI agents.
- Dashboard: A web-based interface for managing and monitoring workflows.
- Client: A library to facilitate the integration of HITL processes into your AI agents.
Highlights of Phantasm
- ✅ Fully open-source and free to use.
- ✅ Compatible with any AI framework or model.
- ✅ Includes a load balancer (Beta) to distribute requests among multiple approvers.
- ✅ Streamlined web-based dashboard for efficient workflow management.
- ✅ User-friendly client libraries available for popular programming languages.
How Phantasm Operates
Phantasm creates a safeguard for AI-driven tasks by introducing an approval layer. This is crucial in scenarios where AI agents might not always be 100% accurate, particularly when the consequences of mistakes can be significant. For instance, when automating calendar scheduling, Phantasm ensures meetings are correctly configured and scheduled before final booking.
Workflow Overview:
- Your AI agent sends an approval request to Phantasm's server.
- The request is relayed to a human approver.
- The approver reviews the request and makes a decision to approve or reject.
- Phantasm communicates the decision back to the AI agent.
In addition, Phantasm allows approvers to modify action parameters before approval, enhancing the workflow efficiency by correcting minor errors.
Best Practices for Using Phantasm
To effectively utilize Phantasm in production, consider the following:
-
Error Handling & Fallback Mechanism: Implement strategies to manage scenarios where approvers are unavailable or requests are rejected. Options include timeouts, fallback actions, and notifications to approvers via email.
-
Parameters & Context: When sending requests, include relevant parameters and context to aid approvers in making informed decisions. Use structured formats like objects or dictionaries for parameters and ensure thorough documentation accompanies requests.
Contributing to Phantasm
Your support is crucial for our community's growth! Star the repository and share it with others to help us reach a broader audience. If you're interested in enhancing the project through code, design, or documentation, please refer to our Contributing Guidelines. Together, we can make Phantasm a robust tool for AI workflows.
A Note of Caution
As Phantasm continues to develop, we encourage users to provide feedback to help us enhance its features and security. Use at your own risk, and stay tuned for improvements!