Area: AI and Intelligent Automation

Problem: A character with a consistent personality, usable memory, and natural conversation is an architecture problem before it is a prompt problem. Most implementations weld personality, memory, and transport together, so nothing can be replaced without touching everything else.

Approach: We designed a hexagonal-architecture bot with strict module boundaries: personality definition, memory, conversation handling, and API integration are separate. No module is longer than 200 lines, which keeps it readable and testable.

Technical details:

  • Hexagonal architecture with explicit ports and adapters
  • Agentic memory with recall, storage, and context management
  • Personalisation inferred from conversation history
  • OpenAI Responses API (GPT-5.4, 1M token context)
  • Personality definitions swapped at runtime
  • Conversation state persisted in Redis
  • FastAPI server with WebSocket streaming
  • 293 lines of architecture documentation

Outcome: Several personalities run on one core. Adding one means writing a module, not editing the system. A restart does not lose the conversation.

Stack: Python, FastAPI, Redis, OpenAI API, WebSocket, Hexagonal Architecture

Page rendered by SimpleCMS — powered by the Plugin System.

Sandboxed plugin ran for projects/character-ai-bot and read “Character AI Bot” through the mediated API.