v0 OpenAI Proxy

Local OpenAI-compatible proxy for Vercel v0 chats.

It exposes an OpenAI-style edge for clients, persists local conversation mappings in SQLite, and serves a lightweight dashboard from the same server.

What It Does

• Accepts POST /v1/chat/completions
• Exposes GET /v1/models and GET /v1/models/:model
• Uses v0 POST /chats and POST /chats/{chatId}/messages upstream
• Stores local conversation state in SQLite
• Serves a dashboard at / with:
  • Overview
  • Usage
  • Conversations

Important Behavior

• If incoming history matches a known local prefix hash, the proxy continues the existing upstream chatId.
• If incoming history does not match, the proxy creates a new upstream chat.
• On that fallback create, the proxy now forwards prior conversation history to v0 by serializing earlier turns into the new chat's system context and sending the current user turn as the live message.
• The Conversations dashboard shows a merged transcript view. When an upstream fallback chat only contains the latest turn pair, the dashboard fills in the earlier turns from the proxy's local history snapshot so you can still inspect the full local conversation.

Install

npm install
npm --prefix web install

Configure

Copy .env.template to .env and set real values:

V0_API_KEY=your_v0_api_key_here
V0_API_BASE_URL=https://api.v0.dev/v1
HOST=127.0.0.1
PORT=3000
OPENAI_API_KEY=local-dev-openai-key
CHAT_STATE_FILE=.data/chat-state.db

Notes:

• V0_API_KEY is required.
• In the current implementation, clients calling the local proxy should send Authorization: Bearer $OPENAI_API_KEY.
• Use HOST=0.0.0.0 if you want to reach the proxy from another device on your LAN.

Database

Run migrations when setting up a new DB file or after adding migrations:

npm run db:migrate

You do not need to run migrations before every start.

Run

Backend and bundled frontend:

npm run dev

This starts the Fastify server on HOST:PORT and serves the built frontend from web/dist.

If you change frontend code and are not running a separate Vite dev server, rebuild the frontend bundle:

npm --prefix web run build

Optional manual frontend dev server:

npm --prefix web run dev

That runs Vite separately, usually on http://localhost:5173.

Client Base URL

For OpenAI-compatible clients:

• Base URL: http://localhost:3000/v1
• Chat completions: http://localhost:3000/v1/chat/completions

If using another device on the network, include the scheme:

• http://192.168.x.x:3000/v1

Models

Current local model list:

• v0-auto
• v0-mini
• v0-pro
• v0-max
• v0-max-fast

Legacy aliases accepted by the proxy:

• gpt-4 -> v0-pro
• gpt-4-turbo -> v0-max
• gpt-3.5-turbo -> v0-mini

Example Request

curl -X POST http://localhost:3000/v1/chat/completions \
  -H "Authorization: Bearer local-dev-openai-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "v0-auto",
    "messages": [
      { "role": "user", "content": "Hello" }
    ]
  }'

Dashboard

Open:

• http://localhost:3000/

Pages:

• Overview for high-level status
• Usage for balance, rate limits, and recent v0 usage events
• Conversations for local conversation list and transcript inspection

API Surface

OpenAI-compatible:

• GET /v1/models
• GET /v1/models/:model
• POST /v1/chat/completions

Internal:

• GET /internal/health
• GET /internal/capacity
• GET /internal/conversations
• GET /internal/conversations/:chatId/messages
• DELETE /internal/conversations/:chatId

Debug-only internal DB endpoints:

• GET /internal/db/schema
• POST /internal/db/query

/internal/db/query is read-only and guarded, but it is still a debug surface, not part of the normal user workflow.

Build

Backend TypeScript build:

npm run build

Frontend bundle build:

npm --prefix web run build

Test

npm test