Turn plain language into a governed planning model.
The Modeler agent builds and edits budgets, forecasts, and scenarios from a plain-language request, working inside definitions your team already governs, on a model reconciled to your actuals.
Margin bridge · draft
Revenue → EBITDA
Works with what you file
One sentence in.
A reconciled model out.
"The Modeler agent works inside the same governed definitions your team already relies on; it doesn't get to invent a new 'revenue' just because a prompt asked nicely."
The Rexfin team
- Filed
- Extracted
- Reconciled
- Cited
- Exported
How it works
Modeler agent overview
Describe what you need
Describe the model in plain language (a hiring scenario, a rolling forecast, a whole new planning surface), from a single metric up to a full application.
01Model a hiring freeze from Q3 through year end.
Reading your headcount and comp definitions…
It maps onto your model
The request is mapped onto metrics, segments, and dimensions already defined in your model, reused, never duplicated, so the plan matches the numbers finance already trusts.
02Which definitions does this touch?
Headcount, comp schedule, and OPEX rollup: all existing.
Formulas get wired in
Formulas, blocks, and dependency links are generated and connected, ready to run through the deterministic compute engine rather than a language model doing arithmetic.
03Building scenario "Hiring freeze Q3"…
14 dependent lines recomputed. Ready to review.
Refine in the same thread
Extend or adjust the model in the same conversation. Structural edits can be tried in a sandboxed branch before they apply broadly, and every change stays logged.
04Now test the same freeze one quarter later.
Branched the scenario: original stays untouched.
Capability overview
Draft. Extend. Explain.
Draft
Build models from a plain-language brief
Structured model generation
Turn a business requirement into structured planning blocks: a single metric or a full planning application.
Formula & dependency wiring
Formulas, blocks, and dimensions get generated and wired to your existing data and business rules automatically.
Import & prep
Upload a CSV or connect a source and have the agent prepare and integrate it directly into the model.
Based on my Revenue Account, and the Rebates and Bad Debt drivers, I built and linked the Revenue Forecast block.
Extend
Evolve models without breaking them
Add scenarios & structure
Layer another planning use case, calculation, or dimension onto a model that's already in production, without starting the build over.
Test changes safely
Try a structural change in a sandboxed branch and validate it before applying it to the model everyone else relies on.
Cleanup & optimization
Surface the parts of a model nobody references anymore, trace what still depends on them, and trim the structure down as the build keeps expanding.
Base model untouched
Explain
Understand what every model does
Query any calculation
Type a question about any line item: where its inputs come from, what's driving the result, and why the number moved this period.
Whole-model map
Trace how a driver, a dimension, and a downstream account connect anywhere in the build, not only within the tab currently open.
Self-writing changelog
Edits are captured with a plain-language note as they happen, so a model can be audited or handed to a new owner without reverse-engineering who changed what.
EBITDA depends on Revenue, COGS, and OPEX: 2 of 3 sourced from reconciled actuals.
Use cases by role
Built for whoever configures the model.
The agent adapts to who's asking; the governed model underneath never changes.
Modelers & FP&A builders
- Build full surfaces faster Draft a complete planning surface from a brief instead of wiring every block by hand.
- Extend without rework Add scenarios or structure on top of what already exists, staying inside the governed model.
- Spend time on judgment, not setup The agent handles repetitive configuration; you focus on assumptions and logic.
Evidence
- Governed definitions
- Deterministic recompute
- Full lineage on every edit
The Rexfin team
CFOs & founders
- Model scenarios without a rebuild Describe an assumption in language and watch the dependent plan recompute.
- Every model stays reconciled Built from actuals already tied to the ledger, not a spreadsheet copy with its own truth.
- One model, every question Board questions and planning changes run against the same governed base.
Evidence
- Governed definitions
- Deterministic recompute
- Full lineage on every edit
The Rexfin team
Controllers & finance leads
- Governed definitions, not shadow copies The agent works inside your metrics library instead of creating a parallel one.
- Lineage on every change Every edit the agent makes is logged and explainable, not a silent overwrite.
- Safer iteration Structural changes can be tested in a branch before they apply to the model everyone relies on.
Evidence
- Governed definitions
- Deterministic recompute
- Full lineage on every edit
The Rexfin team
Accounting-firm partners
- One reconciled model per client Build and adapt planning models across engagements from the same evidence layer.
- Faster answers under review Structure and assumptions stay traceable when a client or auditor asks how a number was built.
- Consistent definitions, engagement to engagement The same metrics library keeps terminology from drifting between clients.
Evidence
- Governed definitions
- Deterministic recompute
- Full lineage on every edit
The Rexfin team
See it in your workflow
A guided look at the Modeler agent.
Walk a live workspace with us instead of reading another feature list.
Further reading
The reliability layer AI needs before it touches your numbers
Why a reconciled financial model, not a smarter model, is the fix for AI that can’t do arithmetic reliably.
Read more GuideAI FP&A automation you can defend in the board room
How driver-based forecasting, variance, and scenarios run on a model AI can operate but never invent.
Read more Product tourA walk through the Rexfin platform
Cited statements, a provenance drawer on every figure, and an export gate that blocks unverified numbers.
Read moreFrequently asked questions
Does the Modeler agent invent its own metrics?
No. It works inside your existing definitions library. Metrics and segments are defined once and reused everywhere, the agent reuses them rather than creating parallel ones.
What actually runs the math?
A deterministic compute engine, not the language model. The agent assembles the model and the formulas; a separate exact engine runs the rollups, ratios, and recalculation.
Can it build a whole planning application, not just one model?
Yes, at a capability level: from a single metric or block up to a broader planning surface, all from natural-language intent.
What happens to my current model when it edits it?
Changes are logged and explained, and structural edits can be tried in a sandboxed branch before they apply broadly, so an existing model doesn’t quietly drift.
How is this different from asking a general AI model to build a spreadsheet?
The agent plans against your governed model and cited actuals, then hands the arithmetic to a deterministic engine; it never asserts a number it can’t trace back to source.
- Filed
- Extracted
- Reconciled
- Cited
- Exported
Describe the model. Let the agent build it inside yours.
The Modeler agent works over the same reconciled model as every other part of Rexfin: no detached scenario, no shadow spreadsheet.