Lipika AI

Anyone doing serious work with language models ends up paying for several of them. ChatGPT for drafting, Claude for long context, Gemini for research, Grok for something else — four subscriptions, four logins, four billing dates, and no single place where the conversation history lives.

The cost is not only money. Switching tools mid-task means losing the thread of what you were doing.

Built the Next.js front end and worked across the FastAPI service layer that fans requests out to the model providers, including streaming responses, conversation persistence and the per-provider error handling.

FastAPI rather than Node for the service tier: the provider SDKs and the tokenisation and routing work around them are Python-native, and bridging that to a JavaScript backend would have meant maintaining a translation layer for no gain.

Responses stream token by token rather than resolving whole. A multi-second wait on a blank screen reads as broken no matter how fast the total response is, and streaming is the difference between a tool that feels alive and one that feels stuck.

Each provider fails differently — rate limits, context-length errors, refusals — so provider-specific failures are normalised into one shape before they reach the interface, and one provider being down degrades to the others rather than breaking the session.

Live at lipikaai.com as a single workspace across ChatGPT, Gemini, Claude and Grok — one login, one bill, one history.