When your Digital Twin answers a question, it doesn’t just pull from its model weights — it grounds the response in the files you uploaded. Citations are the bridge between an answer and the source: they show which segment, which file, and how strong the match was. They turn an AI answer into something you can verify.Documentation Index
Fetch the complete documentation index at: https://docs.praxis-ai.com/llms.txt
Use this file to discover all available pages before exploring further.
Citations exist because trust depends on auditability. Whenever Pria says something specific about your content, you should be one click away from the original passage.
Why citations matter
Grounded answers
Citations tie each claim to the exact segment in your vault. Pria isn’t paraphrasing from memory — it’s quoting source material you can re-read.
Auditability
When an answer goes into a report, contract, or decision, you have a paper trail. The citation is the receipt.
Catches hallucinations
If an answer makes a claim and the cited passage doesn’t actually support it, you know to push back or ask follow-ups. No citation usually means no source.
Speeds up follow-up
Jumping into the cited file is the fastest path to deeper context. You’re already at the right page.
How Pria shows sources
When Knowledge Modes is on (any mode other than Disabled), Pria’s responses can include two layers of citation:1. Inline references
A short citation marker appears next to claims that came from your files — usually a small chip or numbered footnote. The marker says “this part of the answer came from a specific passage you uploaded.”2. Expandable source panel
Below the answer (or in a side panel, depending on your layout), Pria lists every source it consulted while composing the response. Click any entry to expand it and see the matched passage in context.Reading the source panel
Each source entry shows:| Element | What it tells you |
|---|---|
| File name and icon | The document the passage came from. Click to open the full file. |
| Location | Page number (for PDFs / DOCX), timestamp (for audio / video), or segment number (for everything else). |
| Relevance score | A 0–100% value showing how closely the passage matched your question. Higher = stronger match. |
| Snippet | A preview of the matched text, with key terms emphasized. |
| Source label | Whether the match came from semantic similarity (RAG), the knowledge graph (KAG), or both legs combined (fused). |
What the relevance score means
| Score | Interpretation |
|---|---|
| 90–100% | Direct hit. The passage answers the question almost verbatim. |
| 70–89% | Strong match. The passage supports the claim with context. |
| 50–69% | Background context. Related but not directly answering. |
| Below 50% | Usually filtered out — too distant to be useful. |
KAG-augmented and fused citations are common when your Digital Twin needs to connect concepts across files. Pure RAG citations are common when it’s pulling a specific quote.
Opening the original file
Click the file name or icon in any citation to open the File Preview, scrolled to the cited passage. From the preview you can:- Read the surrounding paragraphs for context
- Open the original document in its native viewer (PDF reader, audio player, etc.)
- Jump to other segments in the same file
- Edit the segment if it needs cleanup
- Reprocess the file if you spot a quality issue
When citations are missing
Not every answer carries citations. Common reasons:| Reason | What’s happening | What you can do |
|---|---|---|
| Knowledge mode is Disabled | You turned off retrieval — the LLM answered from its own knowledge alone, no files consulted. | Switch to Normal or KAG Fusion in the Knowledge controls. |
| No relevant match in your vault | Retrieval ran but nothing scored above the minimum relevance threshold. | Try rewording the question, or upload supporting material. |
| The model chose not to cite | The answer didn’t draw directly from any single passage — it was synthesizing across many or applying general reasoning. | Ask the follow-up: “What in my files supports that?” |
| The file is still processing | Recent uploads haven’t finished the pipeline. | Wait for the Included badge, then re-ask. |
| The file is Excluded from RAG | Toggled off in the file action menu. | Open the file’s action menu and Include in RAG. |
| Confidential file, different owner | Pria has access but the segment is locked from your view. | Ask the owner to share, or upload your own copy. |
Showing or hiding citation details
You can control how aggressively Pria surfaces citations from your Instance Settings: The built-in control is the Display RAG/KAG Search Details toggle in Instance Settings. When on, retrieved passages appear inline alongside answers; when off, the answer stands alone and you can still expand the source panel on demand.Cross-checking — best practice for high-stakes answers
For anything going into a deliverable, contract, decision, or compliance review, treat citations as a starting point — not the finish line.Open the cited file
Click into the cited passage. Read it in its native context — the paragraph before, the paragraph after, the section heading.
Verify the claim is actually supported
Does the passage say what Pria summarized it as saying? Or did Pria stretch the interpretation?
Check for missing context
Look at the source panel for other consulted segments. If a critical caveat lives in a segment Pria didn’t directly cite, you want to know.
Ask follow-up questions
“What other passages in [file] discuss this?” or “Is there anything in my vault that contradicts that?” — Pria can search for tension.
Related
- Knowledge Modes — when retrieval runs and which legs contribute
- Searching Inside Your Files — content search across the vault
- IP Vault — vault overview and the RAG/KAG retrieval modes
- Reasoning & Thinking — how Pria’s reasoning interacts with cited sources