alice/ai Multi-Agent Operating System for Small Business

Challenge pressure

Small-business owners lose their week to admin and operations spread across the tools they run the business on, and an assistant that can only answer questions cannot do any of that work.

The situation

alice/ai, formerly PraxisAI, sells an AI operating system to small businesses: the owner sets the direction and specialist agents do the work, from a personal chief of staff to finance and marketing agents. That promise only holds if the agents can act inside the tools the business already uses, and if the product reaches owners on whichever device they work from.

What we built

A FastAPI backend of 51 specialised controllers on AWS ECS Fargate, with a LangGraph Supervisor orchestrating sub-agents over more than 70 MCP tool integrations, retrieval on Pinecone and FAISS, tracing in LangSmith, and GPT-4 and Gemini behind the agents. The React 19 and Vite frontend, on Material-UI v7, Redux Toolkit and React Router 7, streams answers over server-sent events and ships as a PWA and, through Capacitor 7, as native iOS and Android apps. AWS Cognito via Amplify handles multi-tenant sign-in, Stripe handles billing, and 14 DynamoDB tables hold the data. MCP connections to Cliniko and Nookal bring healthcare practice-management records and scheduling into the same assistant.

What changed

Owners work with one supervisor that can act across more than 70 connected tools instead of switching between them, on the web or in the native apps, and the team maintains a single codebase for every platform.

Constraints and decisions

The supervisor routes to specialised sub-agents instead of handing one agent every tool, so each agent chooses from a short, relevant toolset even as integrations pass 70. Native apps come from the same React code through Capacitor, which keeps web, iOS and Android on one release cycle. CI/CD runs on GitHub Actions with OIDC-based AWS access and separate DEV and PROD pipelines, so no long-lived cloud keys sit in the pipeline.

Technology stack

LangGraphLangSmithMCPPineconeFAISSGPT-4GeminiFastAPIAWS ECS FargateReact 19CapacitorAWS CognitoDynamoDBStripe