Ingestion - Private Gpt

Documents are ingested through /v1/artifacts/ingest. Once ingested they are chunked, embedded, and stored in the vector store for retrieval.

Ingest a file

bash

curl -X POST http://localhost:8080/v1/artifacts/ingest \
  -F "file=@/path/to/document.pdf"

Target a specific collection:

bash

curl -X POST http://localhost:8080/v1/artifacts/ingest \
  -F "file=@/path/to/document.pdf" \
  -F "collection=my-collection"

List ingested documents

bash

curl "http://localhost:8080/v1/artifacts/list?collection=my-collection"

Delete a document

bash

curl -X POST http://localhost:8080/v1/artifacts/delete \
  -H "Content-Type: application/json" \
  -d '{"collection": "my-collection", "artifact": "<artifact-id>"}'

Wipe all local data

bash

make wipe

<Warning> This deletes everything under `PGPT_HOME/local_data/` (default `~/.local/share/private-gpt/local_data/`) including the vector store. It cannot be undone. </Warning>

Bulk local ingestion

To ingest an entire folder from the command line, enable local ingestion in your settings:

yaml

data:
  local_ingestion:
    enabled: true
    allow_ingest_from: ["*"]

Then run:

bash

make ingest /path/to/folder

Watch mode (re-ingest on file changes):

bash

make ingest /path/to/folder -- --watch

Supported file formats

PrivateGPT handles plain text natively. The following formats are also supported with built-in parsers:

.pdf · .docx · .pptx · .ppt · .pptm · .hwp · .epub · .md · .csv · .json · .ipynb · .mbox · .jpg · .jpeg · .png · .mp3 · .mp4

Any other file type is read as plain text.