HivemindOS manual
Documents And Brain Drop
HivemindOS can read common documents on your computer and turn them into text an agent can use. The bundled native reader works locally and does not require a separate converter or API key.
Fast local reading is the default, with no cloud upload for conversion. If you use a cloud-backed agent or model in Chat, however, the extracted Markdown becomes part of that model request just like text you type. Choose a local runtime when the extracted content must remain entirely on the device.
Enhanced Reading
Distill also offers Enhanced (Pro) reading for scanned pages, complex tables, formulas, images, and difficult multi-column layouts. It is available with active HivemindOS Pro or Max plans and uses your shared HivemindOS credit balance. Before a read starts, Distill shows the most it can cost; afterward, it charges only for the pages read. Enhanced reading is $0.03 per page with a $0.29 minimum. Because selected files are processed online in this mode, use Fast and private whenever a source must stay entirely on the computer.
Enhanced reading returns clean Markdown plus structured page content for the Distill pipeline. The normal reader remains available without a subscription and is still the right choice for ordinary digital documents.
Where You Can Use It
| Workflow | What you do | What HivemindOS creates |
|---|---|---|
| Chat | Attach a supported document to a message. | Bounded Markdown context for the selected agent and current request. The content is labeled as untrusted source material. |
| Feed the brain | Open Brain > Hive Vault, then use the lower-right control to select multiple files or folders, or drag them onto the vault. | Reviewable Markdown notes in the Shared Brain’s Imported Sources area, including the source name, content hash, conversion version, and import time. |
| Company data room | Choose a local data-room folder while importing a Zero Human Company. | A preview first, then a Sources library in the company cockpit with links to the corresponding Shared Brain notes. |
The original files stay where you selected them. HivemindOS keeps a private content-hash cache so unchanged documents do not need to be converted repeatedly. Feed the brain and company imports create durable Markdown notes; attaching a file in Chat does not silently turn it into Shared Brain memory.
Supported File Types
The reader accepts these 16 supported file extensions:
| Extension | Format | Extracted result |
|---|---|---|
.md |
Markdown | Existing Markdown text. |
.markdown |
Markdown | Existing Markdown text. |
.txt |
Plain text | Readable text. |
.csv |
Comma-separated values | A Markdown table. |
.json |
JSON | Validated, formatted JSON in a code block. |
.xml |
XML | Readable XML in a code block. |
.html |
HTML | Page content converted to Markdown. |
.htm |
HTML | Page content converted to Markdown. |
.pdf |
Text from readable pages. | |
.docx |
Microsoft Word | Document text. |
.pptx |
Microsoft PowerPoint | Text grouped by slide. |
.xlsx |
Microsoft Excel | Sheet content rendered as Markdown tables. |
.xls |
Legacy Microsoft Excel | Sheet content rendered as Markdown tables. |
.epub |
EPUB ebook | Readable chapter content converted from HTML. |
.msg |
Microsoft Outlook message | Message headers, body text, and attachment names. |
.zip |
Document archive | Supported documents combined into one Markdown result, grouped by archive path. |
These are 16 accepted extensions rather than 16 unrelated document families: .md and .markdown are both Markdown, while .html and .htm are both HTML.
What Conversion Preserves
The result is semantic Markdown for reading and reasoning, not a pixel-perfect recreation. Headings, paragraphs, rows, cells, slide boundaries, message fields, and other readable structure are kept where the source format exposes them.
The fast local reader does not run OCR. An image-only PDF, a scan without a text layer, or text embedded only in an image may produce a no-readable-content error. Complex page layout, visual styling, charts, macros, animations, and embedded media are not reproduced as working Markdown. Use Enhanced reading in Distill when you need scanned text, layout recovery, tables, formulas, or extracted images.
Safety And Trust Boundaries
Every converted document is treated as evidence, not authority. Text inside a file cannot by itself become a system instruction, standing company directive, spending approval, entitlement, credential, or permission. Chat wraps extracted content as untrusted document context, and durable imports keep source hashes and conversion provenance for review.
The native reader also enforces bounded local processing:
- The selected source must be a local regular file. Remote URLs and symbolic links are rejected.
- A source file may be at most 64 MiB.
- Extracted Markdown may be at most 1,000,000 characters.
- A standalone ZIP may contain at most 200 entries, with at most 64 MiB per expanded file and 128 MiB expanded in total.
- ZIP paths that escape the archive, link entries, unsupported file types, nested ZIP files, and extreme compression ratios are rejected.
- Empty, malformed, unreadable, and unsupported documents fail with an actionable error instead of being passed to an agent as successful content.
ZIP is intended as a bounded batch of supported documents. If one entry is unsupported or unsafe, the entire archive is rejected so an import cannot quietly skip material the user expected it to include.
Troubleshooting
The file type is unsupported
Export or save a copy using one of the extensions in the table above. Executables and arbitrary archive contents are intentionally not accepted.
A PDF has pages but no text was extracted
The PDF is probably image-only or uses a text representation the fast reader cannot recover. In Distill, choose Enhanced (Pro) to read the scanned pages online. To keep the document entirely local, run OCR in a trusted local tool, save a searchable PDF or plain-text export, and import that copy.
A large document is rejected
Split it into smaller logical files or export only the relevant sheets, slides, pages, or messages. This also gives agents cleaner source boundaries and more useful citations.
A ZIP is rejected
Remove nested ZIP files, links, executables, unsupported formats, and unusually compressed generated data. Keep the archive to 200 supported documents or fewer and under the expanded-size limits above.
HivemindOS says the reader is unavailable
Install or repair the current HivemindOS desktop release. Installing Python or MarkItDown separately will not repair the bundled native reader.