A decade of lost project files, recovered inside the client's own Microsoft 365.

Years of project records sat in several places at once, in different eras and under different names, and the cloud archive had gaps nobody could list. The bigger worry was privacy: any outside firm that cleans up an archive sees everything in it, from client names to contracts and drawings. The company runs inside Microsoft 365 with Copilot under strict privacy rules, so the work had to happen without a copy leaving the building.

PowerShell tools that run on a standard Windows laptop with no admin rights and no installs, delivered by copy-paste and a USB stick. They fingerprint every file by content, group copies by job, keep the best version and label the others beside it, register every job across every era, quarantine duplicates with a manifest and an undo, and assemble one clean archive checked file by file against its source. The original backup was never written to and the live cloud was never overwritten.

The company's own Copilot may read everything, because it lives inside the company's tenancy. Our AI never sees the files. When a log or a summary has to reach us, a small tool on the client's laptop masks it first with a stable table that works both ways: names become codes on the way out and come back on the way in. Only names the client approved stay readable. Staff names never do. The table never leaves the laptop, and the tool proves the round trip before it writes a line. Model training is off on every Crux Labs AI account.

The recovered set is built, verified and assembled into one clean archive, with a run log of every command, a manifest and an undo for every move, and a trail document for the client. The restore to the cloud archive is in progress with the company's IT provider.

PowerShell on the client's Windows laptop, SHA-256 content fingerprints, NTFS hard links, offline HTML reports, Microsoft 365 with Copilot, a masking tool whose table never leaves the client, and the Claude AI model on masked text only.