Project Knowledge Base

The Knowledge Base keeps the material you collect during an examination with the Project where it belongs. Add non-patent literature, patent documents, office actions, search reports, opinions, and other reference PDFs. Goveda Pro extracts their text once, so you and a connected AI assistant can reuse it without sending the whole PDF to a multimodal model each time.

The Knowledge Base is separate from the application's filing documents. Application documents drive the examination workflow; Knowledge Base sources are examiner-supplied reference material.

Capabilities

Keep reference PDFs with the Project

Drag several PDFs onto the Knowledge Base page or choose them from your computer. Goveda Pro uploads each file, extracts its text and images, and keeps the result scoped to that Project. Processing status is shown for each source, from upload through completion.

After a source is added, you can:

  • Give it a clearer title while retaining the original filename
  • Retry an interrupted upload or extraction failure when Retry is available
  • Delete it from the Project when it is no longer needed

Review extracted content against the original

Open a source after its status becomes Completed. The review page places the extracted content beside the original PDF, so you can check the result against the source rather than trusting the extraction blindly.

You can move to a PDF page directly, scroll through the extracted content as one continuous document, and use the same keyword highlights as the Project's patent review pages to find and review matching terms. Extracted tables, formulas, headings, and images are preserved where the source and extraction allow it.

Copy or download the extracted result

  • Copy Markdown copies the complete extracted document as Markdown. If it contains images, their links remain accessible for 24 hours.
  • Copy image link copies a 24-hour link for one extracted image.
  • Download ZIP downloads the Markdown together with any extracted images, so the bundle remains usable without temporary image links.

For a large document, retrieving only the passages you need through MCP is usually more efficient than pasting the complete Markdown into a conversation.

Use the Knowledge Base from Claude or ChatGPT

After you connect Goveda Pro to Claude or ChatGPT, your assistant can:

  • List the sources in a Project and check whether they are ready
  • Search every completed source, or selected sources, for relevant passages or an exact phrase
  • Read a selected page range or limited portion with its PDF page context
  • Retrieve an extracted image when it is needed to understand the passage

The search and retrieval tools return source titles and PDF page numbers, so the assistant can keep the answer traceable to the original material. The tools provide source content; the assistant performs the requested summary or analysis.

For example, you can ask:

Search this Project's Knowledge Base for passages about adaptive robot safety, then summarize the most relevant passages with their source titles and PDF pages.

See Connect to Claude or Connect to ChatGPT to add Goveda Pro to your assistant. The Tool reference lists the exact Knowledge Base tools.

How to use it

Open the Project's Knowledge Base

Open a Project and select Knowledge Base in the left sidebar.

Add the reference material

Drag PDF files anywhere on the page, or select Choose PDF files or Upload PDFs. Each PDF can be up to 200 MB, and you can select several at once.

Wait for completion

You can leave the page while extraction runs. A source is ready for review, search, copy, download, and MCP retrieval when its status is Completed.

Review or retrieve what you need

Select the completed source to compare the extracted content with the original PDF. Alternatively, ask your connected assistant to search the Project's Knowledge Base and read the relevant passages.

Current limitations

  • PDF only. Other formats, including Word, PowerPoint, and spreadsheets, cannot be uploaded yet. Each PDF must be 200 MB or smaller.
  • Extraction is not guaranteed to be exact. Scans, unusual reading order, handwriting, complex tables, and formulas can be misread or reformatted. Check important passages against the original PDF before relying on them.
  • Only completed sources are readable. Content becomes available only when the source reaches Completed. When Retry is available, retry from the web app; otherwise, delete the source and upload a corrected or replacement PDF.
  • Knowledge Base material is not used automatically. Adding a source does not change the Project's application, prior-art search, claim comparison, or report. You or your assistant must choose to search or read the material.
  • MCP management is read-only. A connected assistant can list, search, and read Knowledge Base content, but uploading, renaming, retrying, downloading, and deleting remain web-app actions.
  • Assistant retrieval is deliberately bounded. Long results are returned in parts, and the assistant may make several calls to finish reading a range or document. The retrieval tools do not summarize the content themselves.
  • Image handling depends on the assistant. Some Claude or ChatGPT clients cannot display image pixels returned by a connected tool. In that case, the assistant may ask to open the image link in a browser before it expires.
  • Copied image links expire after 24 hours. Copy the Markdown or image link again to renew access. Images included in a downloaded ZIP do not depend on those temporary links.