simula

A local-first engine for generating and inhabiting worlds and personas from your own materials.

What is simula?

simula is a thin harness that turns your own texts into something you can walk through and talk to. It:

takes your materials (books, texts),
distills a blueprint from them — the spine of a world or a persona (mostly pointers into the material plus a tiny summary), and
runs an interactive, stateful, memory-bearing experience in which the LLM proposes structured changes while the engine holds the truth.

One engine, two blueprint types (world | persona), one unified entity model — the Simulacrum. The name carries a question, not a claim: is a fashioned mind or world real? (PKD)

Why it’s different

Local-first. Runs against a local llama.cpp server by default, with hard-constrained output via GBNF grammars. Your materials and your generation can stay on your machine.
Always portable. The same engine runs against any OpenAI-compatible endpoint with json_schema constrained output.
The engine holds the truth. The model only proposes deltas; the engine validates and applies them against authoritative state. This is what keeps long sessions coherent.
No corpus shipped, ever. You bring your own books; simula extracts texture, not text.

Status

Early alpha (Phase 0). The core is a skeleton; the packaging, CLI, and design are in place. See the Design and Build phases pages for the road ahead, and Principles for the empirically derived lessons that drive every decision.

Install

pip install simula

simula --version
simula init     # create a workspace
simula where    # print the workspace path