Back to catalog
VT Chat
MITSelf-hosted AI chat interface with multi-model BYOK routing, local Ollama support, and browser-only IndexedDB storage.
Open-source alternative to — compare it against:
Overview
VT Chat is a self-hosted web interface for local and commercial language models that isolates conversation history inside the user's browser. Built on Next.js and IndexedDB, its zero-server-storage model ensures that prompts and API credentials never persist on backend database servers. DevOps teams deploy the containerized application to provide engineers with a shared multi-model portal while maintaining strict data boundaries.
Why VT Chat instead of TypingMind
TypingMind charges per-seat commercial licenses for a closed-source web client, leaving organizations dependent on proprietary updates and external hosting. VT Chat replaces commercial licensing with an open-source Next.js application run through Docker Compose, routing prompts directly to upstream providers or local inference servers. Both platforms support Bring Your Own Key configurations across OpenAI, Anthropic Claude, Google Gemini, and xAI, but VT Chat allows engineering teams to inspect the full source code and host the interface inside a private perimeter.
The operational tradeoff balances turnkey convenience against infrastructure control. TypingMind offers hosted cloud synchronization across devices without server maintenance. VT Chat shifts operational ownership to internal teams: engineers configure PostgreSQL for Better Auth sessions, manage environment variables, and provision API credentials. Because VT Chat stores all conversation trees exclusively in client IndexedDB, operations teams avoid the maintenance burden of running a persistent chat database or backing up sensitive conversational data.
What it does well
- Multi-provider model routing: Connects to OpenAI, Anthropic Claude, Google Gemini, and xAI, letting developers switch models mid-conversation while retaining context.
- Local inference integration: Connects with local runners like Ollama and LM Studio over HTTP endpoints, enabling private offline workflows with zero token expenditure.
- Browser-local data isolation: Uses Dexie.js and IndexedDB to keep chats, prompts, and API keys on the client device without persisting messages to backend databases.
- Containerized deployment: Packages the Next.js 15 app, React 19 interface, and Bun runtime into Docker Compose for predictable deployment on internal servers.
- Transparent reasoning inspection: Implements reasoning token support via the Vercel AI SDK, giving developers clear visibility into model thinking stages.
