AgentKit is a starter kit for building constrained agent applications using Next.js, FastAPI, and Langchain. It targets developers looking to rapidly prototype and deploy production-ready, chat-based AI agents with a focus on reliability and user experience. The kit provides a modular, configurable tech stack and a reactive UI for features like streaming, data visualization, and action status.

How It Works

AgentKit addresses agent reliability by constraining execution paths through pre-defined "Action Plans." These plans consist of linear or parallel tool executions, guided by a Meta Agent that selects the most suitable plan for a given user prompt. This approach leverages human domain expertise to steer agents, reducing unexpected behavior. Intermediary outputs from each step in an Action Plan can be streamed to the UI, enhancing user experience by providing real-time progress and data.

Quick Start & Requirements

• Install/Run: docker-compose -f docker-compose.yml up -d (from the root directory after cloning and setting .env variables).
• Prerequisites: Docker, OpenAI API Key, OpenAI Organization ID.
• Setup: Estimated minutes to hours depending on Docker build times and API key setup.
• Docs: Hosted Documentation

Highlighted Details

• Tech stack: Next.js 14, FastAPI, Langchain/Langsmith, NextAuth, Celery/Redis, PostgreSQL with pgvector.
• UI: React/Next.js chat interface with streaming, table/code rendering, and agent action status.
• Reliability: Configurable routing architecture via Action Plans and a Meta Agent.
• Scalability: Includes Queue Management, Auth, Caching, and Monitoring components.

Maintenance & Community

• Active contributors listed in README.
• Open to contributions via PRs.

Licensing & Compatibility

• MIT License.
• Permissive for commercial use and closed-source linking.

Limitations & Caveats

The project is a starter kit; enterprise-grade security functionalities should be added for production deployments, particularly concerning risks like prompt injection.