ethanplusai

before (rip)

AI tools bind context directly to a model (GPT, Claude, Gemini, etc) which makes switching tools unproductive and collaboration on ideas non existent. We kept running into this internally, so we built a system where context is stored in a structured, model-agnostic layer and translated into model-specific prompts at runtime.

The core idea is treating modes as execution layers rather than the source of truth. This lets us switch models, share context across a team or project, and preserve context always.

Still early, but curious how others here think about context management across models, especially for collaborative workflows.