Desktop items, Downloads, project folders, and collaboration tools keep growing. Manual file hunting interrupts demos, reporting, and coding flow.
01 / Case File
Perri Spot
From desktop clutter to one-sentence work retrieval
A useful AI entrance should not force a person to remember paths, filenames, or which window was just open.
Live proofTrained through T1-T12 real prompts, with confirmation flows, recent work memory, and diagnostic logs.
Start with CLI, then connect the desktop window. Train intent recognition, candidate ranking, confirmed opening, organization suggestions, and failure review with real prompts.
Keep Perri’s desktop clean without losing speed during demos, reporting, or multi-project switching.
Desktop items, Downloads, project folders, and collaboration tools keep growing. Manual file hunting interrupts demos, reporting, and coding flow.
Start with CLI, then connect the desktop window. Train intent recognition, candidate ranking, confirmed opening, organization suggestions, and failure review with real prompts.
Keep Perri’s desktop clean without losing speed during demos, reporting, or multi-project switching.
01 / Evidence checklist
Evidence checklist
T1-T12 real prompt training
GUI acceptance tests and failure screenshots
Structured logs and one-click diagnostics
Confirmation-first organization suggestions
Workflow
Workflow
Speak naturally
“Find the file I showed my boss” or “open today’s budget spreadsheet.”
Rank candidates
Use time, path, file type, semantics, and recent work memory.
Wait for confirmation
Risky actions produce reviewable plans before anything moves.
Guardrails
Guardrails
- No deleting, moving, or overwriting project files by default.
- Downloads, Documents, and Desktop can be indexed while dependencies and build outputs are excluded.
- Failures become training samples instead of confident wrong answers.
What it proves
What it proves
- The desktop can gradually clear without losing access.
- Demo file retrieval becomes one sentence instead of manual browsing.
- Failure samples feed a review loop for continuous training.
Next motion