Knowledge

Empower AI agents with trained, unstructured dialogue.

Train the AI agent on the content it uses to generate replies. Training sources are separate from the agent's persona and behaviour — those are configured in the Agent tab. See Manage AI agents.

Add a training source

Select one of the three buttons at the top of the Knowledge base table.

Upload file

Supported file types: csv, docx, html, json, md, pdf, pptx, txt · Maximum file size: 5 MB

1. Select Upload file.

2. Enter a Name for the source.

3. Drag and drop a file or select Browse files.

4. Select Train.

Add URL

1. Select Add URL.

2. Enter a Name and the full URL (must start with https://).

3. Choose the scope: URL webpage trains only on the page at that URL. All webpages crawls and trains on all pages linked from that URL.

4. Optionally enable Automatically retrain every and select an interval: 12 hours, 1 day, 1 week, or 1 month.

5. Select Train.

If the URL points directly to a PDF, CSV, or JSON file rather than a webpage, Proto downloads the file (up to 10 MB) and trains on it as a file source instead of scraping it as a webpage — the same as if it had been added through Upload file. The scope and Automatically retrain every settings above don't apply in that case, since a downloaded file trains once, like an uploaded one. Any other link — an ordinary webpage, an unsupported file type, a file larger than 10 MB, or a URL Proto can't reach — is trained as a webpage as described above.

For URL webpage sources, if a site's bot protection blocks the first attempt, Proto automatically retries through alternative access methods before marking the source as failed. This doesn't apply to All webpages crawls.

Paste snippet

1. Select Paste snippet.

2. Enter a Name.

3. Paste or type the text in the Text snippet field.

4. Select Train.

---

Manage training sources

The knowledge base table lists all sources with name, last updated date, and status. Filter by Search, Type, and Status. A link added through Add URL that resolved to a file (see above) is listed and filterable as a file source, not a URL source.

Status	Meaning
Pending	Queued for training
Loading	Training in progress
Success	Trained and active
Fail	Training failed — remove and try again

To delete a source, select its row and use the delete option.

---

View source details

Select any source from the knowledge base table to open its details panel. The fields shown depend on the source type.

URL

Field	Description
Name	Editable display name.
URL	The trained page, opens in a new tab.
Status	See the status table above.
Last updated	Date of the last successful training or refresh.
Last updated by	The person who created the source, or who last triggered a refresh.

Select the refresh icon at the top of the panel to retrain on the URL immediately instead of waiting for the next scheduled refresh. It's disabled while the source is still pending or loading, and again after use until the panel is reopened.

File

Field	Description
Name	Editable display name.
File	The file name, with a download button for the original file.
URL	The link the file came from. Shown only when the file was created from a URL that resolved directly to a downloadable file (see Add URL above) — an uploaded file has no URL row.
Status	See the status table above.
Created	Date the source was added.
Created by	The person who added the source.

Snippet

Field	Description
Name	Editable display name.
Status	See the status table above.
Created	Date the source was added.
Created by	The person who added the source.

The panel also shows the snippet's text, which can be edited inline. Editing it queues the snippet for retraining.

Below the fields above, the panel also shows a short status message: for Pending or Loading, that the content is still being processed; for Success, that it processed successfully; for Fail, a source-specific error (for example, a prompt to check the URL, file, or snippet and try again).

---

Related

Manage AI agentsLLM processingExample system prompt