Walkthrough
The story
How a .deepcell file gets built, act by act — the claim, the assumptions under it, the calculations that produce it, and the history of every change.
DeepCell Blog
Notes on conclusions that keep their assumptions — and the agents that write them.
Featured
Benchmark
CEO-Bench
One agent, $1M starting cash, 500 simulated days. See the run, then open the model it built.
All posts
What happens when AI opens your model
The number an AI hands back is only as trustworthy as the way it touched your file. Three ways it can — and why only one of them keeps the model alive.
Three ways an agent can touch a model — and the one it reaches for
openpyxl on a binary .xlsx, raw .deepcell XML, or the deepcell CLI. Same task, three very different affordances. Here's why the agent picks the CLI.
Why a new file format
Better AI on Excel can't fix what's structurally wrong with the file. So we built a new one.
Show your work
The cell says 78.4% — but why? How a .deepcell keeps the reasoning behind every number, long after the chat that produced it is gone.
Round-tripping with Excel
Values flow both ways. Structure travels one direction at a time. Here's why that's the right tradeoff.
Bring your Excel model into DeepCell in 10 minutes
Two paths, one non-destructive import, and an honest note about formula conversion.
Anatomy of a .deepcell
Eight plainly labeled sections. Open one in any editor, read it like prose — and leave whenever you want.
How an analyst and Claude share a .deepcell
The file is the shared surface. The agent reads, edits, and proposes; the analyst signs.