EraEvo is not a model — it's the engineering system around models. It analyses intent, decomposes it into tasks, runs a four-agent debate (Architect / Developer / QA / Reviewer) over a strategic decision, builds a dependency graph of files to generate, generates them in topological order with semantic context, validates the AST, builds, runs the self-healing loop on failure, generates tests, runs them, records everything, and feeds the result back into a strategy optimiser so the next run is smarter.
Cloud GPT-4o handles the heavy reasoning. Local Ollama (phi3, mistral) handles everything that doesn't need the big model. A model router and cost meter sit between them, so you don't pay GPT prices for tasks a 7B model can do.