Platform
One engine. Six ways in.
Everything below runs on the ClimRisk desktop app against the local engine instance — the same build our own advisory team opens for every engagement. Hand it an asset registry and it resolves coordinates, pulls real spatial and track-record data, and runs the full NGFS scenario suite in minutes. What comes back still goes through an advisor who can defend the number, not a download link.
Inside the App
What you actually open.
Six tools, each built indigenously and run against the same stress parameters underneath — not six vendors stitched together. The chat agent among them runs with no API key by default, and an optional language model that never leaves your own machine. Here is what each one does, with a real sample run shown where one exists.
AI Risk Analyst
The entry point for most engagements. Point it at a company name, a single site, or a full asset list, and it runs the same pipeline every time: site physical hazards — cyclone, river and coastal flood with local defences — into the carbon policies actually in force, into the NGFS transition paths, out to a Monte Carlo loss distribution and a list of actions. Three ways in, one method underneath.
Three Input Modes
A company name searched against the global heavy-industry asset register (power, cement, steel, aluminium, refining, chemicals, mining); one site by coordinates, sector and activity; or a bulk asset list (.xlsx/.csv) with its own template.
Financials Are Optional, Never Assumed
Without an asset value, money figures come back zero; without emissions, carbon cost is left out. Add commissioning year, revenue, EBITDA, debt and covenant terms and the same run adds EBITDA-at-risk, leverage, credit-rating drift and impairment — the Analyst states what's missing rather than filling it with a sector average.
Saved, Not Disposable
Every run is kept under Recent Assessments and reopens without re-running — so a number you defend to a credit committee this quarter is still the same number next quarter, not a fresh roll of the dice.
Physical Risk Report
A pure engineering read on one site: what the hazard maps say, what that does to the structures actually standing there, and what a named protection measure leaves behind. No financials — this is the report you hand to the plant manager, not the CFO.
Structure By Structure, Not Site-Wide
A 2D hydraulic solver routes mapped flood depths over real terrain and timed water arrival per structure — in one sample run across 64 modelled structures, water reached the first affected tank about 95 minutes into the rise, well before its floor sat clear of the surge.
Wind Against Design Rating, Not a Generic Threshold
Each component's gust loading at its own height is checked against its design wind rating — the same sample run flagged a silo taking 91.2 m/s at 30 m, 5.09 kPa, 166% of its 55 m/s design wind under the 1-in-100-year cyclone.
Protection Scenarios, Quantified
Raise equipment 2.6 m, or build a 3.8 m flood barrier — each option is re-run through the same solver and comes back as "0 inundated, 4 wet" or "1 inundated, 0 wet," not a qualitative before/after.
Asset Twin
The real site, modelled in 3D from its own footprints and terrain — the environment the Physical Risk Report is generated from. You can step through a flood or cyclone replay, compare it against what actually happened at the site since 1980, and test a protection overlay before it's built.
Built From the Site's Own Geometry
Structures are placed on their real footprints over local terrain, not a generic block model — so a flood routed over the twin reaches the buildings that are actually in its path, in the order water would actually reach them.
Validated Against the Historical Record
A tropical-storm track archive back to 1980 and an all-cause wind reanalysis to 1950 are replayed against the twin so a hazard scenario can be checked against what the site has already lived through, not only what a model projects.
A Protection Overlay You Can Test, Not Just Describe
A flood wall, raised equipment or a drainage change is added to the twin and re-run through the same solver — so "what remains" after a mitigation is a number from the model, not an assumption in a slide.
Story Reports
Physical, transition, financial, LCA/ISO 14067, carbon audit and CBAM — delivered as a narrative for a named reader, not a spreadsheet. Each one runs a 500-draw Monte Carlo loss distribution at a 7% discount rate, states the mean and the tail separately, and closes with a "Method and limits" section that says what the model does not cover.
The Mean and the Tail, Both Stated
A sample Financial Risk story on a refinery asset put the expected loss to 2050 at $128.8m (25.8% of a $500m asset value) under a delayed transition — and the 1-in-20 outcome, the number that actually tests a covenant, at $347.4m.
Reader-Segmented, Not One-Size-Fits-All
The same run is reframed for who's reading it: the lender gets the 95th percentile and expected shortfall, the insurer gets the direct-damage share against single-year return periods, the investor gets the spread across scenarios and what it implies for the plan.
Honest About the Edge of the Model
Every report ends with its own limits stated in writing — in the sample above: direct physical damage only, business interruption excluded, a screening-grade model pending owner validation of asset value and defence standards. Nothing is quietly rounded up.
ClimRisk Agent
A resident agent that lives inside the engine itself, not bolted on as a third-party chat widget. As of October 2026 it runs with no API key at all: by default it answers from the sources the engine has learned, assembled rather than generated by an outside vendor, and an optional free model (Ollama, about 2 GB, no account) can be added so it writes more fluent answers — entirely on your own machine, never a cloud call. It can build a twin, run a scenario, research a company's live hazard exposure, monitor a watchlist, and draft the content a risk or IR team needs once the numbers are in.
An Autopilot Inbox
The agent now works in the background while the app is simply open — scanning live hazards, re-running watched sites against the 3D twin, drafting posts — and reports back on request: "what did you find while I was away?"
Four Content Types From One Assessment
A finished run drafts into a LinkedIn article, an investor memo, a case study or regulatory commentary — same underlying numbers, four different readers.
A Watchlist It Actually Monitors
Companies added to a watchlist are checked for material changes without a rerun being requested by hand, with "which companies in my watchlist carry the highest transition risk" answerable directly in chat.
Model Lab
Ask it to build a hazard model and it does the work a modeller would: picks the physics, pulls terrain and reanalysis data, runs the model, puts it through verification tests, checks it against the historical record where one exists, and writes up the method and the result in plain language — across twelve hazard types, from flood inundation and fire weather to earthquake recurrence and storm surge, at any coordinate on request.
Verified Before It's Trusted
A sample flood study at 22.2645, 91.805 passed 7 of 7 internal verification tests in 33 seconds and still reported itself as "verified only" — it had no gauge observations to calibrate or validate against at that site, and said so rather than implying more confidence than the data supports.
Run At 37 Cities, Failures Published
A 222-study sweep across every inhabited climate zone came back 153 of 222 validated or validated with a small, real drift — 69% — with the rest, including one bug found and fixed mid-sweep, written up in full on the Updates page, not quietly dropped.
Some "Failures" Are Climate Signal
At several cities the rainfall or fire hindcast failed because the recent record is measurably higher than the early record predicted — the hazard is intensifying. The engine reports that as the finding, not an error to hide.
How It Reaches You
Built into how we work with you.
The platform above is never sold as a standalone download. It sits underneath one of three engagement models, each ending with an advisor, not a login.
Advisory Retainer
Embedded climate risk intelligence for private equity, asset managers, and lenders underwriting large transactions.
Embedded Due Diligence
Plugged directly into your deal workflow. A TCFD disclosure, an SFDR PAI pack, an EU Taxonomy screen, and a physical risk VaR come back as one engagement, not a subscription.
White Label
Financial institutions license the engine under their own brand for client reporting, with our team running the analysis behind it.
Enterprise API
Feed a coordinate or a portfolio file, get Capital-at-Risk, scenario loss, and disclosure-ready output back programmatically — the same engine, the same validation record, integrated directly into your own risk systems at whatever volume your book requires.