Think it through.
You bring the question. Simmis builds a living model of your domain — your team, your data, your constraints — tests scenarios, and shows you what matters.
Agents handle the modeling. You make the call.
Ask
Type a question, upload files, or connect your existing tools. Start from scratch or fork a shared model.
Explore
Agents build models, test scenarios, and surface what you couldn't see before.
Decide
Inspect the reasoning. Compare outcomes. Act with confidence.
What happens behind the question
You type a question. Here's what the system does with it.
Your knowledge, alive
Describe your domain in conversation — or connect what you already have. Agents structure it into linked, queryable data. Unlike a wiki, this model never goes stale: it's used for every scenario, so drift gets caught.
Every scenario, tested
Ask "what if" and get an answer. The system branches your model, applies the change, compares outcomes. Testing a decision costs nothing — forking a million rows is a pointer flip.
Gets smarter over time
The system observes what actually happens and refines its predictions. Active inference means better answers the more you use it — it builds expectations and updates them against reality.
They do the modeling
Persistent AI teams code, analyze, verify, and cross-validate in sandboxed, versioned environments. They ingest your data, ground models in statistics, and adapt shared templates to your reality. You review and decide.
Proofs, not guesses
Machine-checked theorems from Lean 4's proof kernel certify numerical algorithms. When the system says a result is correct, it can show you the proof. 210k+ Mathlib theorems available.
Scales to real complexity
Million-agent simulations, differential equations, deep learning — compiled to SIMD and GPU compute shaders. The system handles problems too large to spreadsheet.
You already simulate. You just do it in your head.
Every decision you make is a mental simulation — predicting futures, weighing trade-offs, modeling constraints. The trouble is that mental simulations are invisible. They can't be shared, inspected, or improved. When they're wrong, there's no record of why. Your organization's reasoning starts from scratch every time.
Simmis makes that reasoning visible. You ask a question — agents structure your knowledge into a model to answer it. That model persists, grows, and stays grounded because it's actively used. Organization is the consequence, not the prerequisite. Knowledge that participates in inference can't silently rot.
The platform compounds like GitHub does for code. Fork a model someone else built, adapt a template, start from shared knowledge. The system observes what actually happens and refines its predictions — the longer you use it, the sharper it gets. Read the full vision →
Open infrastructure
The Simmis stack is open source. Each layer composes through copy-on-write branching — fork anything, anytime, at zero cost.
| Component | Role | |
|---|---|---|
| Datahike | Immutable database with git-like branching and time travel | |
| Stratum | SIMD-accelerated columnar SQL with copy-on-write snapshots | |
| Raster | GPU-compiled scientific computing, deep learning, and ABM | |
| Ansatz | Lean 4 proof kernel — verified numerical code on the JVM | |
| Spindel | Reactive runtime with O(1) copy-on-write forking | |
| Dvergr | Persistent AI agent teams with sandboxed execution |
From the blog
Visual essays on simulation, modeling, and the infrastructure behind it.