ChatAvatars was a multi-tenant platform that let an organization build a shared set of AI assistants once, instead of every employee writing their own prompts and setup from scratch. Administrators configured each assistant’s instructions, model settings, and PDF knowledge base, then controlled which members could access it.
Architecture
Most of the engineering effort went into the platform around the chat interface: organizations, roles, access control, billing, and per-organization configuration and documents. WorkOS managed authentication, organizations, roles, and plan-based entitlements, and Stripe handled billing.
Stack
- Next.js, tRPC, and TanStack Query
- Neon Postgres with Drizzle ORM
- WorkOS AuthKit and Stripe
- Vercel AI SDK with OpenAI and PDF parsing for knowledge uploads