Platform Overview¶
EpsimoAI is a multi-tenant AI assistant platform built on FastAPI, LangGraph, and LangChain. It supports multiple LLM providers (OpenAI, Anthropic, Groq, DeepSeek, Grok), real-time streaming, document ingestion, and external tool integration via MCP.
Core Concepts¶
| Concept | Description |
|---|---|
| Project | An isolated workspace containing assistants and threads. Each user can have multiple projects. |
| Assistant | A configured AI agent with a specific model, system prompt, tools, and behaviour settings. |
| Thread | A conversation with an assistant. Maintains message history and context. |
| Run | A single execution of an assistant within a thread — the AI processes the input and generates a response. |
Tech Stack¶
- Backend: Python / FastAPI / LangGraph / LangChain
- Frontend: Next.js (React / TypeScript)
- Database: PostgreSQL (RDS) with pgvector for embeddings
- Deployment: AWS ECS Fargate behind an Application Load Balancer
- Auth: JWT (HS256) for users, per-assistant public tokens for integrations
- Payments: Stripe
- Secrets: AWS Secrets Manager