Modelry

Most AI demos are a text box over an API key. Modelry builds the rest of an AI product — auth, persistence, uploads, usage limits, and billing — and makes streaming, token usage, and credit cost visible to the person using it.

Built the whole app: an authenticated, thread-aware streaming chat route; provider switching between local Ollama and Google Gemini; thread history; Vercel Blob attachments; Better Auth with email/password and GitHub; and a credits wallet with a ledger.

• Credits during streaming — credits are reserved before each model call and settled after it, so cost is accounted for while the response streams.
• Provider switching — one AI_PROVIDER setting in a single module selects Ollama or Google Gemini; unsupported values fail fast.
• Abuse protection — Arcjet guards the chat, upload, and tRPC mutation routes.

• Streaming AI chat on the Vercel AI SDK
• Threads: create, rename, delete, revisit
• Credits wallet and ledger with one-time starter credits
• File attachments on Vercel Blob
• Better Auth: email/password, GitHub OAuth, email verification
• Billing scaffolding with Razorpay plans and subscriptions

• Typed end to end with tRPC 11 and Drizzle on Neon Postgres.
• Kept local development free with Ollama as the default provider; Gemini is a configuration switch.

Live app and open-source repository — no claimed user counts.