Bulk Product Description Generator

Q: Does BaseToolbox upload my product catalog?

No. File parsing, prompts, local-model generation, row storage, claim checks, and export run in your browser. Product contents do not enter BaseToolbox analytics.

Q: Can I pause a large catalog and continue after refresh?

Yes, when persistent browser storage is available. The source file and each completed row are saved independently, so a later run skips completed rows and resumes pending or failed rows.

Q: Does the generated copy use facts from other products?

No. Each request contains only the selected facts for one row and its stable identifier. The full table is not placed in the prompt, which reduces cross-product mixing and keeps the batch reviewable.

Q: Will the local AI model download again after refresh?

Normally no. Verified weights are cached by model version and reused after refresh or browser restart while the runtime instance is rebuilt. Clearing site data, private browsing, or browser storage eviction can require another download.

Generate product titles, bullet points, descriptions, customer-email copy, or listing improvement suggestions from selected CSV, TSV, or XLSX facts. Each row is processed locally, saved independently, checked for risky claims and unsupported numbers, and can be resumed after a pause or refresh.

Saved sessions

Drop a product CSV, TSV, or XLSX file here

The file stays in this browser. Legacy .xls files are not supported.

Local AI runtime

Checking this device

BackendWASM

Device profile—

Adaptive limits—

Writing model

Qwen 0.5B · standard

Transcription model

Whisper Tiny · fast

Files, prompts, and results stay in this browser. Browser storage can still be cleared by you, private mode, or browser cleanup.

Private, browser-local product generation

The source file, selected facts, prompts, generated copy, warnings, and saved batch stay in your browser and are excluded from BaseToolbox analytics. The whole table is never sent to a remote model. Verified local model weights are cached by version.

Current limitations

CSV, TSV, and modern XLSX files are supported; legacy .xls, macros, and desktop-Excel formula recalculation are not.
Small local models can still omit, misread, or invent language. Automated checks flag selected risky terms and numbers but cannot prove every claim, so every row needs human review.
Generation is intentionally row-bounded and sequential. Large catalogs require multiple resumable runs, and slower devices use smaller batches and conservative model settings.
Browser storage can be cleared or reclaimed, and private browsing may not retain sessions, source files, generated rows, or downloaded model weights.

How bulk product copy generation works

Choose trusted product facts

Import a spreadsheet, select a stable identifier, and explicitly choose the columns the model is allowed to use.

Generate and save one row at a time

The shared local model receives one bounded product record, returns constrained JSON, and each row is persisted before the next one starts.

Review flags and export

A deterministic scanner flags selected risky claims and numeric values absent from the source row. Retry individual rows or export all statuses and disclosures to CSV or XLSX.