Meridians Engine
Any text is a world view.
Extract it, query it, simulate where it goes next.
or from a premise
Paste any long-form text and Meridians extracts a typed knowledge graph you can query by meaning, interrogate in-character, simulate forward, and extend with new content — all from the same substrate.
A causally coherent, queryable knowledge structure measured across three force fields: fate (commitments pulling toward resolution), world (the state of entities), and system (the rules of the domain). A novel, a research paper, and a wargame brief are all world views with different weightings.
Integrated Obsidian with a quantitative engine that simulates forward. Notes give you a static graph; Meridians gives you one that's queryable like a database, chattable like a character, and brancheable like Git.
Semantic search with AI-synthesized overviews and citations. Surveys distribute one question across the cast to reveal fault-lines. Interviews go deep on one subject. Every respondent answers in-character from its own continuity.
Yes. Every character holds a private knowledge graph — what they've seen, who they trust, what they believe — and only that gets loaded when you talk to them. Nothing leaks across; the world stays internally consistent.
Yes — the core loop. The engine extracts load-bearing variables, generates a cohort of next-arc scenarios over them, and ranks each with a relative probability. Run them all in parallel; the top becomes the active branch, the rest stay as sister divergences.
Git-like. Branches fork from a parent and share its timeline by reference; only structurally different scenes create new objects. Revise a whole branch through review → verdict → reconstruct to produce a new version. Alternate futures stay grounded in the same root world.
Fate as information gain over thread prediction markets (attention-weighted KL divergence). World and System as graph deltas. Per-scene game theory across 14 strategic axes with continuous ELO ratings. Pacing as Markov transitions. All derived from graph deltas, not prose.
Yes. A Phase Reasoning Graph mines the world's machinery; a Causal Reasoning Graph plans each arc; scenes execute the graph paced by Markov chains from analyzed world views; prose follows beat plans that re-render into prose, screenplay, meta, or simulation formats.
Free and open source. Bring an OpenRouter key for analysis, search, and generation. Optional OpenAI key for embeddings, Replicate key for images. You pay only for tokens — no subscription, no platform fee.