How to Upload Files for Private Assistants
Aumente las capacidades y el alcance de sus propios asistentes
When you create an assistant and set it to private, you gain the ability to upload files in two distinct sections: Code Interpreter and File Search. These files enhance the assistant’s ability to analyze, transform, and search for data when responding to user queries.
Code Interpreter
The Code Interpreter allows you to analyze and transform data directly within the chat. Uploaded files are processed by the assistant to perform operations, computations, or analyses based on their content. For example:
If you upload a Python script (
.py), the assistant can execute it or analyze its content to solve problems.

Supported File Types:
c, cpp, docx, html, java, json, md, pdf, php, pptx, py, rb, tex, txt, css, js, ts, csv, jpeg, jpg, gif, png, tar, xlsx, xml, zip
File Search
The File Search section allows the assistant to use your uploaded files as a knowledge base for answering questions in the chat. These files are indexed, enabling the assistant to retrieve relevant information or data to address your queries.
You can ask the assistant for specific keywords, columns, or data within the files, and it will provide the most relevant results. For example:
If you upload a PDF containing technical specifications, you can ask, “What is the voltage specification for the device?” and the assistant will search the document to provide the answer.

Supported File Types:
c, cpp, docx, html, java, json, md, pdf, php, pptx, py, rb, tex, txt, css, js, ts
RAG: Source Citations
RAG-backed answers now include inline citations and a ranked source list so you can verify where information came from.
What you’ll see
Inline markers like [1], [2] next to facts or passages in the answer.
A collapsible “Sources” list under the answer showing the top 3–5 sources by relevance.
Rank number and title
Clickable link (opens in a new tab)
Source system badge (e.g., SharePoint, Confluence, Azure Blob)
How to use citations effectively
Click the [N] marker or open “Sources” to jump to the referenced document for confirmation.
If multiple sources disagree, prefer higher-ranked items; the top of the list is most relevant.
Use this feature in reviews and audits to document provenance quickly.
If no sources are shown, the answer may not have used your repository context. Re-ask your question and be explicit about the file, folder, or topic to guide retrieval.
RAG: Confidence Indicators
Each RAG-backed answer now displays a confidence badge (High, Medium, Low) based on retrieval signal strength and agreement across sources.
High
Strong agreement across multiple relevant sources. Suitable for operational use with minimal follow-up.
Medium
Some relevant evidence found, but partial coverage or moderate agreement. Spot-check key claims.
Low
Weak matches or conflicting evidence. Treat as tentative — verify using the cited sources.
Guidance by confidence level
High: Proceed; optionally open sources for due diligence.
Medium: Validate critical points via citations before executing decisions.
Low: Click into sources and rephrase your question to narrow the scope or upload clearer reference files.
Low confidence often improves if you specify the exact file, section, or policy date, or reduce the question scope.
Last updated
Was this helpful?