Personal Agent
Multi-agent system with LangGraph for task automation. Supervisor pattern, tool orchestration, NestJS backend with CQRS. Extensible tool system via REST API - add your own integrations through the panel or by asking the agent.
Architect & Full-Stack Developer
Technologies
What I Built
LangGraph Supervisor Pattern
Architecture with central supervisor routing requests to specialized agents: TaskAgent (CRUD tasks, scheduling), SystemAgent (configuration, integrations). Each agent has its own StateGraph with explicit state management.
Extensible Tool Framework (REST)
Framework for dynamically adding tools via REST API. Define an endpoint, parameter schema - and you have a new tool for the agent. You can ask the agent to add integration with any API (e.g., 'add weather API integration').
Real-time Dashboard (React)
React frontend with WebSocket streaming of agent responses. Context API for state management, tanstack-query for cache. Panel for managing REST integrations, conversation history, task queue.
Challenge
I wanted an AI assistant not limited to predefined integrations. Every project has different needs - one uses Notion, another Airtable, third a custom CRM. I needed a system easily extensible with any API.
Solution
LangGraph for agent orchestration with explicit state management. Supervisor analyzes intent and routes to appropriate agent. CQRS in NestJS - separate command/query handlers. Key: REST-based tool framework - define new integrations via endpoint and schema.
Key Decisions
LangGraph over simple chains - explicit state, conditional edges
NestJS CQRS for separation of read/write operations
Prisma with PostgreSQL - typed queries, migrations
WebSocket + Server-Sent Events for different use cases
React Context API - native state management without boilerplate
I type 'add weather API integration and remind me about the meeting'. Agent defines a new REST tool, creates scheduled task in DB. Zero hardcoded integrations - the system extends itself. It's like having a developer-assistant who understands context and builds their own tools.
Kacper Kawczak
Architect & Full-Stack Developer
Results
System handles any REST integrations. I can ask the agent 'add integration with API X' and describe the endpoint - agent creates a new tool. Zero hardcoded integrations - everything extensible via panel or conversation with agent.