Custom Sources
Not everything you need to review lives in PubMed or Semantic Scholar. Custom sources let you bring your own documents into a Literature Review Agent project — proprietary clinical study reports, internal research memos, regulatory submissions, or anything else that isn’t publicly indexed. Once ingested, they’re searchable and extractable alongside the standard databases.
Uploading a ZIP Archive
This is the simplest option for most teams. Bundle your documents into a ZIP and upload it — that’s it.
- In the Custom Sources section, click Upload ZIP.
- Select a ZIP archive containing
.pdf,.txt, or.docxfiles. - The appliance extracts text from each file and indexes it for keyword and semantic search.
- When ingestion is done, the source appears as a selectable option in Step 1 (Search Articles).
Ingesting from an S3 Prefix
If your documents are already in S3, there’s no need to download and re-upload them.
- In the Custom Sources section, click Add S3 Source.
- Enter your bucket name and prefix (e.g.
my-bucket/reports/2024/). - The background worker fetches and ingests all supported files under that prefix automatically.
- Make sure the IAM credentials used by the
workercontainer haves3:GetObjectands3:ListBucketon the target bucket.
Managing Custom Sources
| Action | UI | API |
|---|---|---|
| List sources | Custom Sources page | GET /custom-sources |
| View a source | Select from list | GET /custom-sources/{id} |
| Check storage usage | Sources page header | GET /custom-sources/space |
| Upload ZIP | Upload button | POST /custom-sources/upload |
| Add S3 source | S3 form | POST /custom-sources/s3 |
| Search a source | Search panel | POST /custom-sources/{id}/search |
| Delete a source | Delete icon | DELETE /custom-sources/{id} |
Privacy
Custom-source documents are stored and processed entirely inside the appliance. They’re never sent to any external database or indexing service. The one exception is extraction itself: article text is sent to your configured LLM (SageMaker or DeepLens) for processing — which is the same behaviour as for articles from public databases.