Multi-LLM
A multi-model AI chat platform with real-time streaming across Claude, GPT-4o, Gemini, and Grok — built to compare model responses side by side and explore what each provider does differently.
CloudAgent is a streaming multi-LLM chat interface that lets you send a single prompt to multiple AI providers simultaneously and watch the responses arrive in real time. The goal was to build something that makes model differences tangible — not just in output quality, but in latency, tone, verbosity, and reasoning style.
The project grew out of a practical need: when building AI-powered products, developers constantly ask "which model should I use?" CloudAgent turns that question into an interactive experiment rather than a theoretical debate. You can switch providers mid-conversation, lock in a specific model, or run parallel threads for direct comparison.
Temporal is used for durable workflow orchestration, ensuring that long-running agent tasks survive network interruptions and can be inspected or retried without losing conversation state.
Frontend: Next.js with TypeScript, using the Vercel AI SDK's streaming primitives to pipe tokens directly into the UI as they arrive from each provider. Each model gets its own streaming lane so responses never block each other.
Backend: Node.js API layer that normalizes the wildly different streaming formats from Anthropic, OpenAI, Google, and xAI into a consistent event stream the client can consume without provider-specific logic.
Orchestration: Temporal workflows handle multi-step agent tasks — web research, tool use, and chained completions — with full replay durability so nothing is lost if a step fails partway through.
Infrastructure: Deployed on AWS. Provider keys are managed through environment configuration, making it straightforward to add or rotate access without touching application code.