Copilot Instructions for AgentKit
January 3, 2026 ยท View on GitHub
Project Overview
Multi-agent chat application using LangGraph for orchestration and AG-UI for agent-frontend protocol communication.
Stack:
- Backend: Python, FastAPI, LangGraph, AG-UI, pytest
- Frontend: TypeScript, NextJS, Shadcn UI, AG-UI
Architecture
Backend
- Agent Orchestration: LangGraph multi-agent workflows
- State Management: LangGraph state graphs with conditional routing
- LLM Integration: Ollama provider with
qwen:7bmodel (extensible to other providers) - API Server: FastAPI with CORS, streaming SSE endpoints
- AG-UI Protocol: Real-time agent stream events to front-end app as JSON stream.
- Observability: LangFuse integration (optional), structured logging
- Tools: Search tools, code execution capabilities (Optional)
- See .github/agents/backend.agent.md for detailed patterns
Frontend
-
UI Framework: NextJS + Shadcn UI for modern, accessible component library
-
AG-UI Client: Receives event streams from backend (RUN_STARTED, TEXT_MESSAGE_CHUNK, RUN_FINISHED, etc.)
-
TypeScript: Full type safety with proper interfaces
-
Shadcn UI: Modern, responsive UI design with Shadcn UI components
-
State Management: React hooks for agent state tracking
-
See .github/agents/frontend.agent.md for UI patterns
Development Patterns
Agent Workflow Design
When building LangGraph agents:
- Inherit from
BaseAgentclass with AG-UI integration - Define clear state schemas using
AgentStateTypedDict - Use conditional edges for intelligent routing between agents
- Implement tools as separate classes inheriting from
BaseTool - Emit AG-UI events for frontend visibility (THINKING, EXECUTING, COMPLETE, ERROR)
- Support streaming responses through SSE
LLM Integration
- Use
LLMProviderFactoryto get provider instances - Default: Ollama with
qwen:7bmodel - Extensible to OpenAI, Anthropic, and other providers
- Configure via
config.pyand environment variables
Frontend Integration
When building UI components:
- Use Shadcn UI components for consistent, accessible UI design
- Integrate AG-UI client to receive real-time event streams from backend
- Type all agent states and messages with TypeScript interfaces
- Handle SSE events through AG-UI client for real-time updates
- Connect to backend using typed API client in
services/api.ts - Use Tailwind CSS with Shadcn UI for styling with responsive design
Specialized Agents
This project uses specialized chat agents for focused development:
Implementation Planner (implementation-planner.agent.md)
- Use for: Creating implementation plans, breaking down features
- Outputs: Markdown files in
/.docs/1-implementation-plans/ - Coordinates: Work between Backend and Frontend agents
- Boundaries: Does not implement code directly
Backend Agent (backend.agent.md)
- Use for: LangGraph workflows, agent nodes, state graphs, routing logic, agent tools, FastAPI endpoints
- Tech: Python, LangGraph, FastAPI, Ollama, LangFuse (optional)
- Virtual Env:
/.venv - Boundaries: Does not modify frontend or infrastructure
Frontend Agent (frontend.agent.md)
- Use for: AG-UI components, React UI, agent visualization, TypeScript types
- Tech: React, TypeScript, NextJS, Shadcn UI, AG-UI, Tailwind CSS
- Boundaries: Does not modify backend agent logic or LangGraph workflows
Key Files
- agents.md - Multi-agent architecture overview
- .github/agents/implementation-planner.agent.md - Feature planning guide
- .github/agents/backend.agent.md - LangGraph implementation guide
- .github/agents/frontend.agent.md - AG-UI guide
- .docs/1-implementation-plans/ - Feature implementation plans
When Implementing Features:
- Backend: Design LangGraph workflow, agent nodes, and tools
- AG-UI Protocol: Define communication contract (events, state sync)
- Frontend: Build UI components with AG-UI stream events and Shadcn UI
Knowledge base
- Refer to .docs/1-implementation-plans/ for implementation plans
- Refer to .docs/2-knowledge-base/ for comprehensive knowledge base
Key consideration & practices
- When you finish, always update the knowledge-base documentation in
/.docs/2-knowledge-base/to reflect new patterns or components created in a well-organised manner. Don't create any documentation outside of/.docs/2-knowledge-base/, - Don't create any other guidance outside of the knowledge-base documentation.
- You must follow best practices for building NextJS apps with Shadcn UI and AG-UI integration.
- Always validate architecture decisions
- Always consider refactoring for better code quality and maintainability
- Write meaningful logs
- always use context7 to fetch latest document and code snippets.