Knowledge Library Try Protos

← Home · Knowledge Library

The Knowledge Library is Protos's institutional memory. Every decision, data point, and reference is captured here and linked to the design artifacts that used it — so nothing is ever lost between projects, teams, or people.

Set this up before using the Co-Engineer. The Co-Engineer draws on the Knowledge Library when creating schemas, filling data documents, and answering questions. The richer the library, the more grounded and traceable its output.

What Lives Here

Type	Examples
Academic literature	Papers, standards, patents — uploaded as PDF, DOCX, XLSX, XLS, CSV, TXT, Markdown, JSON, and common image formats up to 100 MB
Internal reports	Experimental summaries, design reviews, test reports
Decisions	Why a parameter value was chosen; why an approach was rejected
Experimental results	Test reports and datasets uploaded as files or captured as text notes
AI-surfaced connections	Co-Engineer surfaces relevant prior work as you design

Source Categories

Every document is automatically classified by where it came from, so you can tell a peer-reviewed paper apart from a forum thread at a glance. Each category has its own icon in the library table:

Category	What it means
Research	Peer-reviewed papers, standards, patents
Article	Articles and other published web writing
Web	General web pages
Forum	Forum and discussion threads
Datasheet	Manufacturer datasheets and spec tables
Uploaded	Files you uploaded directly
Memory	Knowledge notes you (or the Co-Engineer) captured as text
Model Run	Content generated from a model run
Conversation	Knowledge captured from a Co-Engineer conversation
SharePoint	Content the Co-Engineer imported from SharePoint

Protos assigns a category on ingestion, and the owner can change it at any time: open the ⋯ row actions and use Set category (or select several rows to set the category in bulk).

Files you import yourself through Add → From SharePoint are recorded as Uploaded. The SharePoint category applies to imports the Co-Engineer makes on your behalf.

Finding Knowledge

Open Knowledge from the sidebar.
Switch between views with the tabs at the top: List (the default flat, paginated view), Knowledge Graph, and Project Graph — see Traceability and Connections.
In the List view, narrow things down with the category filter chips (Research, Article, Web, …), each showing a count — only categories actually present in your library appear. Combine them with the tag filter, the search box, and a sort dropdown offering Newest first, Oldest first, Name A–Z, or Format.
If documents have been shared with you, Mine and Shared chips also appear so you can switch between your own documents and ones others shared. A marker on each shared row shows it came from someone else.
Use the search box to find items by title, folder path, or tag, then press Enter — it doesn't search document content or notes.
Click any item to see its full content, its source categories and tags, the sources that informed it, and what it is used by. From here the owner can also download the original file (or the extracted text, for conversation-sourced entries), and for entries created from a Co-Engineer conversation you can jump back to that conversation.

Tip: Before starting a new project or design iteration, browse the library for prior experiments and decisions in the same domain. Co-Engineer can also surface relevant entries automatically as you work on the canvas.

Adding to the Library

The Add menu offers four ways in, described below in the order they appear.

Upload a document

Click Add → Upload Document.
Choose a file — PDF, DOCX, Excel, CSV, TXT, images, and more. Up to 100 MB per file.
The title is auto-filled from the filename — edit it if needed.
Click Upload. Protos parses and chunks the content, making it available to the Co-Engineer.
If that file is already in the library, Protos tells you so instead of creating a duplicate, and offers View document to open the copy you already have.

Upload a folder

For bulk ingestion of many files at once:

Click Add → Upload Folder.
Select a folder — Protos processes the files in the background. Batches are capped at 1 GB and 12,000 files. A large selection is split across several background jobs, so one upload can show as several entries in the jobs panel. A file Protos can't parse is reported as a failure, and the rest of the batch is ingested as normal.

Several uploads can be in progress at once; Protos works through them as capacity allows. There's a limit on how much can be uploading at once, so a large upload may be asked to wait. Protos tells you when that happens, and you can start the same files again once the ones ahead of them finish.

Import from SharePoint

Click Add → From SharePoint. The Browse SharePoint dialog opens.
If your Microsoft account isn't connected yet, connect it here — or from Integrations in the sidebar. In a tenant where individual users can't grant access themselves, a Microsoft administrator approves Protos once for the whole organisation, and everyone connects normally afterwards. Your IT team can also limit that approval to named sites and folders.
Browse sites, then document libraries and folders, or switch to Search to find a file by name within the site you're in. Tick anything you want to bring in: individual files, whole folders including everything nested inside them, or an entire document library. One selection can mix all three within a site — moving to a different site clears it. Up to 400 items can be ticked individually; a folder counts as one item however much it holds, so tick the folder when you want more than that.
Click Import. Protos scans your selection, then ingests on the server, reporting how many files were added, how many were already in the library, and how many failed.
Closing the dialog only stops the progress display — the import keeps running on the server. Reload the page to see files that landed after you closed it.

A SharePoint import has its own, much larger limits: up to 12,000 files and about 22 GB in one go. A selection bigger than a single ingestion batch is split across several background jobs. Individual files above the per-file size limit are reported as failures, and unsupported file types count toward the total and are reported the same way.

Add a knowledge note

To capture a decision, insight, or observation as text:

Click Add → Add Knowledge.
Enter a Title and the Content.
Click Save Knowledge.

What Protos reads from a file

Text is extracted, chunked, and embedded so the Co-Engineer can search it and quote it back with a source trace.
Tables are read as tables, not as flattened text — in Excel workbooks and in delimited text files (.csv, .tsv). A decimal comma and thousands separators are understood: Protos works out which grammar a column uses from the file itself, column by column, and leaves a column alone rather than guessing when it can't tell. That lets the Co-Engineer pull a specific column and filter rows on a value. It returns the rows themselves: there is no summing or averaging a column. Where a sheet is too ambiguous to read confidently, Protos says so on the document instead of guessing at it, and a file that isn't really a table is read as text.
Formulas in a workbook are read as well: Protos sees each sheet's formulas and which sheets feed which, described in terms of the columns a formula reads rather than as cell addresses. Nothing is recalculated — Protos reports what the file says. Where it isn't confident it has read a sheet's columns correctly, formulas are left unlabelled rather than guessed at.
Drawings and scanned PDFs are searchable: a page with no usable text layer is read by a vision model.

Spot-check critical values from large drawing sheets. Very small text on A0/A1-sized pages can be misread.

Attach spreadsheets to the library, not to a chat message. A file attached directly in a chat is read as text — only Knowledge Library documents get the table treatment.

Managing documents and folders

Use the ⋯ menu on a document row to Rename, Set category, Share, or Delete it. Folder rows offer Rename and Delete. From a document's detail page you can also edit its title inline, and edit the content of a knowledge note you created. (Tags are edited from the document's row in the library table — see Tags.)

Traceability and Connections

When the Co-Engineer creates or updates a data document using information from the Knowledge Library, it records which specific chunks of which documents it drew on. This means you can see exactly where a field value came from — not just "the Co-Engineer said so" but the specific source passage, with a citation that links straight back to it.

Field value in data document
  └── Knowledge Library chunk
        └── Original uploaded document (paper, report, note)

Used by

Each document's detail page has a Used by panel — the reverse of a citation. It lists the schemas, data documents, models, and canvases that were built from that document, so you can see the downstream impact of a source before you change or remove it. For a document that came from a Co-Engineer conversation, this panel is instead titled "Modified in this conversation" and lists what was created or edited during that chat.

Graph views

Two graph views let you explore these relationships visually:

Knowledge Graph — how your documents and the assets that cite them connect across the library.
Project Graph — the same connections scoped to a single project.

This chain is preserved permanently — even if team members leave or projects are archived.

Knowledge documents are private to you by default, and can be shared read-only with other people — no one you share with can edit or delete them. There are two ways a document gets shared:

Directly — open a document and use its Share dialog to grant read access to specific people or teams.
Automatically, with the work built on it — when you share a project or an asset (schema, data document, model, or canvas), Protos additively shares the Knowledge Library documents that asset cites, so your collaborator can open and reason over the same sources.

Shared documents appear in the recipient's library under the Shared filter, marked with a "shared with you" indicator. Access is always evaluated live, so revoking a share takes effect immediately. See Collaboration & Sharing for how sharing works across Protos.

Best Practices

Capture decisions as they're made, not retrospectively. The rationale is clearest in the moment and becomes harder to reconstruct over time.
Link papers to specific claims, not just to the paper itself. Trace is only useful if it points to the exact piece of evidence that informed a decision.
Use descriptive titles and tags so documents are easy to find — search matches titles, folder paths, and tags (not document content), and tags also let you group documents across categories.
Keep categories aligned with how you work: adjust a document's category with Set category whenever a different one fits your workflow better, so filters and the graph stay meaningful.
Review the library at project kickoff: search for prior experiments and decisions before starting new work. Don't repeat work that's already been done.