upload

$limai upload <extractionSchemaId> <path>

Upload a single file or an entire folder for data extraction. Supports concurrent uploads, glob filtering, automatic processing, and waiting for results.

Options

--concurrency <n>Number of concurrent uploads for folder uploads(default: 5)

--processQueue files for extraction after upload

--waitWait for all extractions to complete (implies --process)

--output <file>Write results to a JSON file

--glob <pattern>Filter files by glob pattern (folder uploads only)

--allow-duplicatesUpload a file whose content the project already stores as a new document instead of skipping it

--jsonOutput raw JSON

--verboseEnable debug logging

Behavior

Single file: Uploads the file and optionally queues it for extraction.

Folder: Scans the directory for supported files, uploads them concurrently, and reports results.

When --wait is used, the CLI polls until all extraction jobs complete. --output then writes the extracted data for a single-file upload, and the per-file upload summary for a folder.

Each file's content is hashed and matched against the documents the project already stores. A match is skipped: no bytes are uploaded, nothing is queued for extraction, and the exit code stays 0, because a duplicate is not an error.

A single-file upload prints:

Duplicate of existing file <fileId> (<filename>); skipped. Pass --allow-duplicates to upload it as a new document, or run 'limai documents process <extractionSchemaId> <fileId>' to re-extract the existing one.

When the match lives in another model the message names it instead, and drops the limai documents process suggestion — that command would fail here, so --allow-duplicates is the only way forward:

Duplicate of existing file <fileId> (<filename>) in model <modelId>; skipped. That document belongs to another model, so 'limai documents process' on this model would fail; pass --allow-duplicates to upload it into this model as a new document.

With --json it reports {"fileId": "<existing id>", "filename": "...", "isDuplicate": true, "bytesUploaded": 0, "skipped": true, "existingDeploymentId": "<model id or null>", "existingInSameModel": true}.

On a folder upload each file's status can be duplicate, and the per-file JSON entry carries isDuplicate. The summary line reads:

Upload complete: <n> uploaded, <n> queued, <n> duplicates skipped, <n> failed (<n> total)

followed by one line per duplicate:

Duplicate: <filename> matches existing file <fileId>

or, for a match in another model:

Duplicate: <filename> matches existing file <fileId> in model <modelId>

isDuplicate means the document is already stored in the project, not that it has been extracted. To extract the copy that is already there, run limai documents process <extractionSchemaId> <fileId> — which works only when the existing copy lives in the same model. With --allow-duplicates the file uploads as a new document instead and is processed normally when --process is passed.

Examples

Upload a single file

limai upload es_abc123 ./invoice.pdf

Upload and extract immediately

limai upload es_abc123 ./invoice.pdf --process

Upload, extract, and wait for results

limai upload es_abc123 ./invoice.pdf --wait --output results.json

Upload a folder of PDFs

limai upload es_abc123 ./documents/ --glob "*.pdf"

Bulk upload with higher concurrency

limai upload es_abc123 ./documents/ --concurrency 10 --process

Upload folder, process, and wait for all

--output here writes the upload summary — one entry per file with its status, job ID and duplicate information — not the extracted data.

limai upload es_abc123 ./documents/ --wait --output batch-results.json

Upload a second copy of a document the project already stores

limai upload es_abc123 ./invoice.pdf --allow-duplicates --process