Evidence-native modelling
Build a causal case you can defend — and others can follow.
If your work comes down to a causal argument — the drivers, the evidence, the logic connecting a decision to its outcome — Latris is where you build it, justify it, and make it land.
Today that lives in a sketch in Miro, a draft in ChatGPT, citations in a reference manager, and a report in Word — and the reasoning only really exists in your head. Latris pulls it into one place: every link has a reason and a source, AI helps you draft while you stay in control, and what comes out is clear enough that the people who need to act on it can see exactly why.
Latris is open — sign in with your email at the link above, no invite needed. Questions first? Get in touch.
Runs in any modern browser — nothing to install.
Good thinking gets lost in the gaps.
Building the model is only half the job. The other half — assembling the evidence behind every link, recording what you considered and ruled out, turning it into something a client or reviewer can trust — is slower, and it's where the work falls apart. Provenance scatters across five tools, and six months later no one can say why a particular arrow is on the page.
AI makes this sharper, not softer: the easier it gets to generate a model, the more every relationship needs a reason to exist. And the tools built for this are either a decade out of date or so specialised that only their author can drive them — so the findings never really leave your screen.
Where Latris fits
Built for work like this.
Latris suits work where the reasoning is causal and the stakes make it worth defending — whether you're a consultant, a researcher, or an in-house team. It's built around two kinds of work:
Systems thinking — environment, health & policy
Environment & natural resources
What's driving change in a catchment or ecosystem — and which interventions actually move the needle?
Public health & wellbeing
Mapping the causes behind an outcome to target where action will help most.
Policy & programs
The causal logic behind a program or policy, made explicit, sourced, and reviewable.
Risk & safety
Natural hazards & risk
How the factors behind a hazard combine, where the risk is most sensitive, and what the evidence says.
Infrastructure & engineering
Reliability and failure reasoning, with the evidence behind each link kept on the record.
Safety & assurance
Claims, arguments, and evidence laid out so a reviewer or regulator can follow — and challenge — every step.
Every model type shares one evidence, provenance, and review layer underneath. System dynamics — stocks, flows, unit-checked equations, and simulation — is already in the open-source engine and CLI; the visual stock-and-flow editor in the app is on the way.
Bayesian networks: set up states and conditional-probability tables, then click a node to enter evidence and watch the belief bars update across the network. Sensitivity analysis and learning tables from data are in the open-source engine and CLI today.
Don't see your field or your method? The underlying approach is general — if your reasoning is causal, it probably fits. Other formalisms (decision analysis, process models, and more) aren't on the roadmap right now — if your work needs one, say so. Get in touch and tell me what you're working on.
Why Latris
Defensible
Every link has a reason and a source. Provenance, evidence, and challenge are built into the model — so when someone asks "why is this arrow here?", there's an answer, not a shrug.
Clear
Your model becomes something others can follow — a clean diagram, a sourced report, a plain-language walkthrough — without you in the room to narrate it. Honest about what it can't say, because the uncertainty is measured, not spun.
Fast
AI drafts and reviews while you stay in control. One tool instead of five. Get to a credible, sourced model in an afternoon — fast enough that you actually do the defensible version.
Built for the moment it leaves your desk.
Findings that travel
Most modelling tools assume you'll be standing next to the screen, explaining. Latris assumes you won't. The same versioned model produces what each audience needs, and the reasoning travels with it — click any claim and trace it back to the evidence behind it.
Board-ready diagram
A clean, grid-aligned picture of the system — presentation-quality without a trip through another tool.
Sourced report
The written case, with every claim tied to its evidence and assumptions made explicit.
Plain-language story
A walkthrough for the stakeholder who just needs the "why" — no jargon, no prior knowledge assumed.
One model, three audiences — all from the same source of truth.
For the people who own the decision
When the model isn't yours, but the risk is.
If you commission, sign off, or inherit model-backed work, you carry the risk without having built the thing. Latris gives you the answers you actually need: what assumptions drove this result, who changed what and when, which version backed the decision, and whether another team could pick it up in eighteen months. The full history is kept automatically — Latris becomes the institutional memory around the analysis, not just the place it was drawn.
A note from the founder
I'm Shannon Bengtson — I design and build Latris. I've spent enough time around people who build the causal case behind a decision to watch the same thing happen again and again: careful reasoning gets built in a hurry, the evidence behind it lives in scattered documents, and six months later no one can say why a particular arrow is on the page.
Latris is the tool I wished existed — structure, provenance, and AI assistance that keep pace with the work, so the result holds up when someone asks hard questions, and travels to the people who need to act on it. It's early, and the people I'm bringing on now genuinely shape where it goes. If that resonates, I'd love to hear from you.
Connect on LinkedInOpen core
Built on an open backend.
The modelling engine behind Latris is open source (Apache-2.0) — a deterministic, git-native CLI and API designed to be driven from a terminal or by an agent: schema-validated models, provenance on every mutation, evidence and challenge as first-class data. The hosted app is a browser GUI built directly on this same engine — same git history, same schema, same CLI underneath. Both are open to anyone today.
No API key? If you work with an AI assistant in your terminal, it can answer Latris's structured prompts itself — your agent does the drafting, you stay the approver, and every step stays on the record.
pip install "latris[server,data]"
Tell me what you're building
Latris is open — sign in and start modelling, no invite needed. If you want to flag what you're working on or hear about new releases, leave your email below.
No spam — occasional release notes only. Prefer a direct line? [email protected] or LinkedIn.