Your model, refreshed from the filings, without an AI touching your formulas.
OpenHelm pulls the latest actuals from SEC filings, maps them to your Excel model, and hands back a reviewable diff with a citation on every cell. The agent proposes; your analyst approves. Formula cells are never overwritten, and every run is backed up first.
now
Already in production with
The control room
Built for analysts who will never trust a black box near their model.
It proposes, you approve
The agent does 90% of the fetching, parsing and cell-mapping, then halts. Nothing is written to your model until an analyst reviews the diff and marks it approved. "Review-ready model refresh", never "AI that edits your files".
Formula-protection guarantee
On inspection every cell is classified as a hardcoded input or a formula. Formula cells are read-only. If an approved edit now targets a formula, the apply job halts, flags an integrity error and skips the cell rather than break your math.
A citation on every number
Each proposed change carries the exact cell address, an old → new diff, a confidence rating and a direct link to the SEC EDGAR filing and XBRL tag it came from. The analyst never has to guess where a number originated.
Automated, timestamped backups
Before any write, the apply job saves a timestamped copy of the workbook (Tesla_Model_Backup_2026Q1.xlsx) to the parent folder. Styles, macros and untouched cells are preserved by a non-destructive writer.
Works with the storage you already use
A uniform file toolset reads and writes Google Drive (narrow drive.file scope via the Picker, no CASA review), OneDrive / SharePoint via Microsoft Graph, and AWS S3 via your IAM keys. Keep using Excel exactly as you do today.
A memory that learns your mappings
When an analyst approves or adjusts a mapping, OpenHelm remembers it in a model-map table. Low-risk, settled mappings can be flagged auto-apply; everything else keeps the human in the loop.
Job 1 · Propose
Discovery and extraction, written to a review ledger.
- 1
Secure retrieval
The target .xlsx is downloaded from your repo (Drive, OneDrive or S3) into an isolated execution sandbox. Nothing runs on your machine.
- 2
Workbook inspection
openpyxl and pandas inventory every sheet, label and reference, classifying each cell as a hardcoded input or a read-only formula.
- 3
Filing extraction
The SEC EDGAR company-facts API is queried with a compliant user-agent for the latest raw XBRL financial facts on the ticker.
- 4
Semantic alignment
Raw concepts (e.g. us-gaap:RevenueFromContractWithCustomer…) are mapped to your custom workbook labels, reconciling unit scales and reporting periods.
- 5
Divergence analysis
Reported numbers are compared with what is already in the cells. Matches are ignored; only genuine divergences become proposals.
- 6
Proposed-changes ledger + email
Each proposal (cell address, diff, confidence, citation, rationale) is written to an interactive ledger, then a summary email lands: "Apple reports Q1 revenue $119.5B. 14 updates proposed for review."
Job 2 · Apply
The committal step is deliberately boring, and safe.
When an analyst marks changes approved, the apply job re-downloads the live file, backs it up, verifies each target is still a hardcoded input, writes only those cells, and uploads, flagging every row applied.
| Manual data entry | Raw AI that edits files | OpenHelm | |
|---|---|---|---|
| Who writes the cell | Analyst, by hand | The model, unsupervised | Analyst approves an agent-drafted diff |
| Formula safety | Depends on the analyst | Can clobber formulas | Formula cells are read-only, integrity-checked |
| Provenance | Re-keyed from the filing | None | Direct SEC EDGAR + XBRL citation per cell |
| Backups | If you remember | Rarely | Automatic, timestamped, every run |
| Where data lives | Your storage | Vendor cloud | Pulled into a single-use sandbox, then destroyed |
What clients say
OpenHelm basically gave us an entire marketing department overnight. This is what it feels like to punch above your weight!
Dr Thom Van Every
Founder · Smoothie Wars
Common questions about model maintenance
More use cases
Other jobs OpenHelm runs.
Hedge funds & asset managers
Thesis-aware morning briefing
Overnight filings and news, scored against the thesis you hold per ticker.
SEO & GEO growth
SEO & GEO keyword research
Search and generative-engine demand mapped to the terms worth winning.
SEO & GEO growth
Publish-ready content to your stack
Briefed, drafted and pushed straight to your repo, WordPress or Shopify.
SEO & GEO growth
Continuous improvement over time
Live pages monitored and refreshed as rankings and AI answers move.
Twenty minutes
Bring a model and a recent filing. We'll propose the diff live.
Show us one workbook and a ticker that just reported. We'll run the propose job on the call, walk the proposed-changes ledger cell by cell with you, citations and all, and show you exactly where the human stays in control.