Lobster Model Switcher

The Problem

In multi-model assistant setups, the issue is rarely whether a model exists. The friction appears when switching models breaks continuity: a task is underway, a stronger model would help with inference, but the user must re-explain files, goals, and preferences. xiaolongxia-model-switcher addresses this “switch models without breaking memory” workflow by turning model selection into a conversational action.

How It Works

• Seamless switching: ask the assistant in natural language, such as “Lobster, switch to doubao-seed-2.0-code” or “use gpt-4o”, and the session is redirected to the target model.
• Context retention: the skill emphasizes preserving conversation history, long-term memory, daily notes, user preferences, and workspace context, so background does not need to be re-pasted.
• Task-based recommendations: for prompts like “recommend a model for coding”, it suggests candidates based on task type, such as code, writing, long documents, Chinese tasks, or mathematical reasoning.
• Candidate list: the documentation lists Volcano Ark models, OpenRouter models, and free fallback models as a practical reference.

Boundaries

It is best suited to assistant environments that already support multi-model calls, model routing, or an external gateway. The models actually available depend on platform APIs, credentials, quotas, and endpoint limits; if a target model is unavailable, the skill can usually only suggest alternatives or report currently usable options.