Knowledge and Sources
Knowledge keeps copied source files separate from paid Searchable Knowledge, so your employee can use the right material without treating every note as permanent recall.
TL;DR
Knowledge has two layers. Sources copy chosen text files into an employee's Drive without a credit charge, while Searchable Knowledge is a separate paid index for material you intentionally train. Start from Knowledge, chat, or your desktop companion.
How It Works
Sources are original files copied from a folder you approve. They become ordinary Drive documents and stay available for the employee to read when the work needs them. Importing never reads every file into chat or charges credits for a model to summarise it.
Living Knowledge means the normal Drive documents that you or an employee can keep editing over time. They are useful working material, not a hidden memory file.
Searchable Knowledge is the existing paid training area. It is for stable material you deliberately want indexed for automatic retrieval. A Source does not enter that paid path automatically, and a changed file needs a fresh estimate before indexing.
How Employees Find Useful Context
Your employee can draw on four kinds of information. They serve different purposes and do not all run on every request.
| Information | How it helps | What controls it |
|---|---|---|
| Current conversation | Keeps your request and recent discussion in view | The conversation you are working in |
| Conversation memory | Recalls preferences, decisions, and connections between people and projects | Advanced Memory and access to the original conversation |
| Searchable Knowledge | Finds relevant passages in documents and sources you have trained | Your plan, completed training, and source access |
| Files and connected tools | Reads a working document or checks a current record when the task needs it | Available tools, connected accounts, and permissions |
For example, when you ask for a customer reply, Searchable Knowledge can supply the refund policy, memory can supply your preferred tone, and a connected tool can check the current order. These sources help the employee answer your question without loading every document into every conversation.
What Happens When You Train Material
- You choose the files, text, website, or connected content to include and confirm training.
- Sistava extracts readable text and indexes smaller passages. Check the training result before relying on the source.
- On an eligible request, it searches the material you are allowed to use for passages related to your question, including similar wording.
- The selected passages and their source details become context for the employee's answer.
Training makes your material searchable. It does not rewrite the underlying AI model, and it does not guarantee that every relevant passage will be found. For an important detail, ask for the source and review it. Retrain or refresh changed sources; a trained copy is not a live view of the original.
Learn more about conversation memory and Searchable Knowledge.
Add a Source
- Open Knowledge on the Company Dashboard and select Add source. When you ask your employee to import existing knowledge, it opens this drawer directly.
- Pick a familiar location such as Codex, Claude, or OpenClaw, or choose Any folder.
- Your desktop companion opens the native folder picker. Choose one folder.
- Choose Only me or Share with workspace.
- Review the grouped preview, then confirm the one bulk copy.
- Open Files to browse the copied documents in Drive.
What It Can Do
| Action | What happens |
|---|---|
| Import a folder | Copies supported text files into the employee's Drive after your confirmation |
| Use a suggested source | Starts from a recognised assistant location, then still asks you to choose the folder |
| Resync | Opens the picker again and compares the new folder contents with the source history |
| Open files | Takes you to the ordinary Drive files that were copied |
| Share a source | Shares every copied file with workspace readers, or gives selected people read or editor access |
| Make searchable | Selects up to 20 stable copied files, then shows a free estimate before paid indexing |
| Remove a source | Removes the source from the Knowledge list while keeping its retained Drive files |
How to Set It Up
Install and sign in to the desktop companion when you want to import a local folder. The companion is the part that can open your operating system's folder picker. The web app never scans your home folder or receives an absolute local path.
If your employee needs background early in a relationship, it may offer this flow. You can also ask at any time to import a project folder, prior notes, or another assistant's knowledge. A normal folder is valid even if you have never used another assistant.
Tips and Tricks
- Start with one useful folder. Import the project notes, operating docs, or workspace material that will matter for the next work.
- Use resync for changes. Choose the folder again when you want to refresh it. Unchanged files are not copied twice.
- Keep living work in Drive. Edit copied notes or add reports beside them as the project develops.
- Keep focused questions small. If one file answers the question, share that file instead of importing a large folder.
- Treat imported skills as reading. An imported
SKILL.mdis a reference document, not a new employee instruction.
What It Costs
Copying a Source is free. The grouped preview and transfer do not call a model or build embeddings.
Automatic Searchable Knowledge retrieval checks access first. If your plan does not include training, or you have no usable trained content available to the request, that document search is skipped. A paid plan alone does not trigger an empty document search. Trained websites, pasted text, and connected sources count as content even when you have never uploaded a file.
Advanced Memory has its own setting and can use credits without any uploaded documents. Reading files, using connected tools, and generating an answer can also use credits. See Memory for that separate control.
To index a Source, select Make searchable on its card, choose up to 20 stable Drive files, then select Get free estimate. The estimate counts the current file revisions and shows credits before the Confirm paid indexing button appears. Declining leaves your copied Drive files available at no cost.
Is It Safe
Sources are private to the person who imports them by default. The companion scans only the folder you choose and excludes protected, unsupported, oversized, hidden, repository, executable, and escaping symlink files. It sends grouped metadata for the preview, not your local computer path.
Sharing happens from the Source card, not separately on each copied file. Share with workspace gives workspace members read access. You can add individual editors when someone should edit the files or run a resync. Removing a person's source access removes their copied-file access too.
The employee can suggest the flow and open Knowledge, but it cannot choose the folder, scan your device, approve the preview, or read the copied files all at once. It reads individual Drive documents only when they are relevant to work.
Good to Know
- A resync never deletes files already retained in Drive because a later local scan cannot find them.
- Removing a Source hides its source record. It does not delete the copied Drive files.
- Plain text, Markdown, logs, JSON, and similar text material are the right first import. PDFs and Office files use their dedicated training or upload flows.
- Sources and Searchable Knowledge are intentionally different. Copied files stay editable and on demand, while paid training is for stable material you want automatically retrieved.
Frequently Asked Questions
Q: Do I have to select every file?
A: No. Choose one folder in the desktop companion, review the grouped preview, and confirm the bulk copy once.
Q: Can I import from Codex, Claude, or OpenClaw?
A: Yes. Knowledge can suggest those locations when the desktop companion recognises them. You can always choose any other folder instead.
Q: Does importing cost credits?
A: No. Sources copy supported text files into Drive without model or embedding work. Searchable Knowledge is the separate paid feature.
Q: Do employees search an empty knowledge base on every question?
A: No. Document retrieval is skipped when the feature is unavailable or there is no usable trained content. Conversation memory is separate and can still help with earlier discussions when enabled.
Q: What if I change a file after I see the searchable estimate?
A: Sistava stops that run before indexing the changed file and asks you to review a fresh estimate.
Q: Can my employee access my whole computer after this?
A: No. Every import starts with your native folder picker and your grouped confirmation. The employee has no authority to scan other folders.
Q: What happens when I remove a source?
A: The source leaves the Knowledge list, but the copied Drive files remain available until you explicitly remove those files from Drive.