Files & Resources

The Resources drawer is where everything attached to a workspace lives — uploads, downloads, web search results, knowledge, profile, and templates. This guide covers how to upload files, what types are supported, and how the model uses your files when answering questions.

The Resources drawer

Every workspace window has a Resources tab. Open it to see everything attached to this workspace, organised into tabs: Uploads, Downloads, Websearch, Knowledge, Profile, and Templates. This isn't storage for its own sake — everything in the Resources drawer is context that can inform the model's responses. Files you've uploaded, pages that have been fetched during research, and facts from your knowledge base are all drawn on automatically when they're relevant to what you're asking.

Uploading files

Click Upload in the Uploads tab to add a file to the workspace. Files are stored on your node — they don't leave your hardware except as part of an inference request when their content is directly relevant to a question. After upload, the file is processed: text-native files (plain text, Markdown, HTML, CSV, JSON, XML, YAML) are ingested directly into the knowledge base. Binary files — PDFs, Word documents, Excel spreadsheets, and images — go through an extraction step that converts them to readable text first.

Both paths result in the file's content being available to the model when it's relevant. The processing happens automatically in the background after upload; you don't need to do anything once the file is uploaded.

The 10 MB per-file limit applies to all uploads. Files larger than 10 MB cannot be uploaded. For large documents, consider splitting them into smaller files by section or chapter.

Supported file types

Two categories of file are supported: text-native and extractable. Text-native files are read directly — their content is plain text that needs no conversion. Extractable files go through a processing step that converts them to text before they're ingested.

Type Extensions How it's processed
Plain text .txt Read directly
Markdown .md Read directly
HTML .html, .htm Read directly
CSV .csv Read directly
JSON .json Read directly
XML .xml Read directly
YAML .yaml, .yml Read directly
PDF .pdf Text extracted by the extraction service
Word document .docx Text extracted
Excel spreadsheet .xlsx Text extracted
Image .png, .jpg, .jpeg Visual content extracted by the extraction service

Files that don't match any of the supported types are stored but not processed — the model cannot read their content. If you need to reference content from an unsupported file format, copy the relevant text into a plain text or Markdown file and upload that.

How the model uses uploaded files

When you send a question, Varen searches your uploaded files — along with your knowledge base — for content relevant to the question. If a file contains relevant content, it's included in the model's context automatically. You don't need to tell Varen which files are relevant; it works this out from the content of your question.

You can also refer to files by name to bring them in explicitly — "summarise the Q4 report I uploaded" or "what does the lease say about the break clause?" This is useful when the question is specifically about that document and you want to make sure the model focuses on it.

Tip

For dense reference documents — a tenancy agreement, a financial statement, an inspection report, a medical letter — uploading them and then asking specific questions is often more effective than pasting the text into a message. The model can re-read the file in future sessions without you re-uploading it, and you can ask different questions about the same document across multiple sessions.

Downloads

When the model produces a file as part of a response — a formatted document, a spreadsheet, a structured brief, a draft letter — it appears in the Downloads tab of the Resources drawer. Click the filename to download it to your device. Downloads are associated with the workspace and remain available in future sessions, so you can come back and retrieve a document the model produced weeks ago without needing to regenerate it.

Web search

Varen can search the web as part of answering a question. When it does, the pages it retrieves are recorded in the Websearch tab in the Resources drawer. Each entry shows the URL, the search query that triggered the fetch, and when it was retrieved. This is your audit trail of what the model consulted during research.

If a response surprises you or includes information you didn't expect, checking the Websearch tab can tell you whether a fetched page contributed to it — and if so, which one. This is especially useful when the model draws on current information (pricing, policy details, news) that might have changed since a page was retrieved.

Templates

Templates are reusable structured prompts scoped to a workspace. They're designed for starting a common type of conversation without retyping the same setup each time — a weekly financial review format, a standard property inspection checklist, a recurring meeting-prep structure. Templates are coming soon.