Read, write and search the wiki

Each space shows the content/ folder of its Git repository as a wiki.

The tree is on the left and the page on the right.
The space’s Settings (the gear next to its name) give the sync state (Git repository) and the list of members (Members).

Read

Markdown: rendered with headings, tables, checkboxes and highlighted code.
- Relative links ([guide](../infra/proxmox.md)) lead to the target page. A link that starts with / starts from the root of content/.
- Relative images are displayed.
Folder: its README.md or its index.md serves as an introduction, followed by the list of files.
Other files (see Files other than Markdown):
- PDF: the browser’s viewer (or an Open the PDF link on a phone or tablet);
- images: displayed;
- HTML pages: rendered in an isolated frame;
- Office documents: download, with their extracted text;
- text and code files: syntax highlighting;
- other types: download.
Pages linking here: the Markdown pages that link to the page shown.
See also: related pages, found without searching (see below).
History: each change, with its author and the detail of the changed lines.

Below Pages linking here (and, on mobile, after the content), the See also panel shows up to 8 pages close to the page shown: their title, path and the reason.

same topic: meaning, when the space has semantic search (embeddings); two pages that talk about the same thing in different words find each other.
shared words: rare words in common (accents and word forms folded, as in search; common words hardly count).
cites this page: it links to the page shown; linked from this page: the page shown links to it; shared links: they cite the same pages, or are cited by the same ones.
same folder: a slight bonus, never enough on its own.

The panel loads after the page, without slowing it down, and stays hidden when nothing is close enough: better nothing than noise. PDFs, Office documents and images appear through their extracted text; a file without text, never.

not linked yet flags a related page that the page shown does not link to. A contributor can click Suggest a link: a note asking to link the two pages is dropped in the inbox, and the nightly tidy-up will add the link (nothing is written to the wiki right away). The suggestion is made only once; while it waits, the panel shows Suggestion pending.

Assistants get the same list (MCP tool related, REST GET …/files/{path}/related), and the tidy-up agent receives it when it reads each note: enough to complete an existing page rather than duplicate it.

Files other than Markdown

Markdown first, and all your files. Pages are Markdown, and only they have links, backlinks, an outline, verification and related pages. But a space holds any file, versioned in Git like a page (history, move, delete), and Synaptane gets the most out of each one:

Format	In the wiki	Search	AI assistants (MCP)
PDF	Built-in viewer	Text layer, or OCR for scanned pages	Read the extracted text
Word, PowerPoint, Excel (`.docx`, `.pptx`, `.xlsx`, and the old `.doc`, `.ppt`, `.xls`), OpenDocument (`.odt`, `.odp`, `.ods`)	Download, extracted text below	Extracted text	Read the extracted text
Images (PNG, JPEG, GIF, WebP, BMP, AVIF, SVG)	Displayed	Text in the image, by OCR (not SVG or AVIF)	Read the OCR text
HTML (`.html`, `.htm`)	Rendered in an isolated frame, source one click away	Words	Read and write
Other text files (CSV, JSON, YAML, code, logs…)	Syntax highlighting	Words	Read and write
Anything else (archives, videos…)	Download	File name	Listed, no text

On spaces where semantic search is enabled, every text in this table is also searchable by meaning.

PDF: shown in the browser’s PDF viewer, inside the page. Below 700 px wide or on a touch screen, where that viewer often stays blank, an Open the PDF link above it opens the file in the browser’s own reader. Its text is extracted page by page; a page without text (a scan) is read by OCR with Tesseract, in the languages set by the administrator (French and English by default). Very long documents are read up to 500 pages, and up to 100 scanned pages.
Word, PowerPoint, Excel and OpenDocument: no preview in the browser, a Download link. Their text is extracted and shown under the link, in Extracted text (used by search): the body of a Word document, the text of the slides (not the speaker notes), the text and values of spreadsheet cells: for a formula, its result as saved in the file, not the formula itself.
Images: displayed in the page, and in Markdown pages that embed them (![rack](rack.png)). The text they contain (a diagram, a screenshot, a photo of a label) is read by OCR, unless the administrator turned it off (ocr_images). An SVG is displayed as an image, its scripts never run; opened on its own, it is downloaded.
HTML pages: rendered in a frame isolated from Synaptane (sandbox): no access to your session or to Synaptane, no network requests from scripts, no form submission. View source shows the code, highlighted. It is never rendered on its own at Synaptane’s address: opened outside its wiki page, it is downloaded. Search finds words of the source.
Text files (CSV, JSON, YAML, logs, scripts, configuration…): highlighted according to their extension, editable like a page, searchable.
Other files: stored and versioned, offered for download, found by their name.

The text of a new document becomes searchable a few seconds to a few minutes after it is added, the time for the worker to extract it.

Adding files:

web UI: Add in a folder (contributors), 25 MB per file;
from an AI assistant: as attachments of a note (inbox_add, base64, up to 10 files of 25 MB each), with the folder in hint; the nightly tidy-up files them near the page they relate to and links them;
REST API: PUT …/files/{path} with content_base64, or attachments of an inbox note;
Git: commit them under content/ and push. Files over 64 MB are kept but their text is not extracted.

Old Office formats: the text of .doc, .xls and .ppt files (binary formats from before 2007) is extracted too: .doc and .xls by catdoc, included in the Docker image, .ppt by Synaptane itself. They are shown like the others: a download link, the extracted text below. What is read, and the limits:

Word (.doc): Word 97 to 2003, and also Word 6 and 95; an RTF file named .doc is read too. One line per paragraph, table cells separated by tabs.
Excel (.xls): Excel 97 to 2003, and Excel 5 and 95. One line per row, cells separated by tabs, a blank line between sheets. A date is written as a date (2019-04-20) only in one of Excel’s built-in formats; in a custom format, and in files saved by LibreOffice, it shows as a number (days since 1900).
PowerPoint (.ppt): PowerPoint 97 to 2003, and files saved in that format by LibreOffice (other exporters are untested). Titles, bodies, free text boxes, tables (one line per row, cells separated by tabs) and speaker notes, slide after slide (a blank line between slides). Not read: PowerPoint 95 files, the text of masters (footers set there), of WordArt, of charts and of embedded objects, comments, and text that is only in a picture.
A file protected by a password, or damaged, has no text: it stays downloadable and is found by its name. For the best search, save these files as .docx, .xlsx or .pptx.

Without catdoc (installation from source without the catdoc package), .doc and .xls files are kept and downloadable, without text, and synaptane doctor warns about it; after installing it, synaptane reindex reads the files already there. .ppt files do not need it.

Write (contributors)

Edit a text page, New page, Add a file (25 MB maximum, any type: see Files other than Markdown), Move (or rename) or Delete this file (⋮ menu).
Each save produces a commit in your name, pushed to the repository at once.
Move keeps links right, in the same commit: the relative links of the moved page are recomputed for its new location, and the pages that linked to it are rewritten (relative links, and [[folder/name]] links that would no longer lead to it). A [[name]] link that still finds it is not touched, and only the link changes: the rest of the page stays as it is. The message shown after the move lists the updated pages. MCP, REST and the tidy-up agent move files the same way.
Each changed page also shows up in the inbox (a Change item, computed from the Git history). The nightly tidy-up will check its location, its links and its consistency with the neighboring pages. A change to spaces or blank lines only is ignored.
A text that seems to contain a secret (password, API key, private key) is refused before it is committed: it would end up on the Git host. Remove it (keep it in a password manager), or check This is not a secret: save anyway if it is a false alarm.
If the page has changed in the meantime (another person, an external push), the save is refused and your text is kept in the form. Reload, then carry over your changes.

Verify a page

To know, when reading a page, whether you can trust it: who answers for it, when it was verified, whether it is due for review. Everything is in the page’s front matter, so in Git:

---
owner: ada            # an account username, or a free-form name
verified: 2026-10-03  # date of the last verification
review_every: 6m      # 90d, 12w, 6m, 1y; absent: the space's interval; never: never
---
# Failover procedure

State of a Markdown page: verified (a date, and the due date not yet passed), to review (due date passed) or never verified. Without an interval, neither its own nor the space’s, a verified page stays verified. Other files have no state.
On the page: a discreet badge in the top bar, “Verified on Oct 3 · Ada” in green, or “To review since …” in amber; nothing for a page never verified. The details (owner, due date, interval, “modified since it was verified”) are at the top of the ⋮ menu, and on hover over the badge. On mobile, the badge shrinks to its icon.
Mark as verified (contributors, ⋮ menu): reread the page, then the button writes today’s date in verified; a page without an owner becomes yours. The same form chooses the owner (an account of the space, or a free-form name) and the interval; Save changes them without dating. It is a commit in your name (Page verified: <path>, or Page tracking updated: <path> when only the owner or the interval changes) that only touches the lines of these three fields: the rest of the text is kept byte for byte, and the block is created at the top if it does not exist. Such a commit does not count as a change for “modified since it was verified”; neither do those written in French by earlier versions (subjects starting with Page vérifiée or Suivi de la page mis à jour), still recognized.
Tidy-up: a commit that only changes these three fields is not a change to review (the content is compared, not the author or the message: the same thing pushed from Git is recognized too). A page changed and then verified still has to be reviewed.
AI agents: the tidy-up agent and assistants (MCP) can neither erase nor change these fields by rewriting a page: they are restored. A page changed by the agent stays verified; the page says “modified since it was verified” (compared by day), and it is up to its owner to reread it.
Lists and search: a green (verified) or amber (to review) mark follows the page name in folders and in results. Search ranks verified, up-to-date pages a little higher (+10%), pages due for review a little lower (−7%): the factor is multiplicative, relevance remains the first criterion.
Space interval: Settings → General, Verified pages: review (1, 3 or 6 months, 1 or 2 years, or never). A page that carries review_every keeps its own.
Reminders: the home page lists, under Waiting, the pages due for review that you own (recognized by your username or your email address). A grouped email goes out at most once a week, on a working day, to each owner of overdue pages (verified address; uncheck Receive emails from the routines of my spaces in My account to stop receiving it). The Consistency check routine receives the list of pages to review.

The purpose of the space can say which pages must be verified, and by whom (most templates suggest it in their “What matters most” section, “Ce qui compte le plus” in French).

Wiki health

The Health tab of a space (gear icon, next to Activity) runs a check without AI, at the latest commit:

broken links, with a suggestion when a file with the same name exists elsewhere;
orphan pages: no other page leads to them (folder README.md and index.md files do not count);
empty or untitled pages (neither a # Title nor a title: in the front matter);
pages to review: verified, but the verification due date has passed (owner and date), and among verified pages, those without an owner;
old pages: not changed for 3, 6, 12 or 24 months, according to Git.

Each line leads to the page, and to Edit for contributors. The same check is served by the REST API (GET /api/v1/organizations/{org}/spaces/{space}/health, parameters months and limit). The tidy-up also runs it at the end of a run, on what it wrote.

Search

Menu field: searches all your spaces.
Field of a space: searches that space only.

It behaves like grep:

case-insensitive and accent-insensitive search (“procedure” finds “procédure”), singular as well as plural (“servers” finds “server”); a word of three letters or fewer only matches a whole word;
each word counts, except stop words (“de”, “la”, “the”…); pages that contain all the words, or the exact phrase, come first;
Regular expression for patterns, for example back\w+ night.

In spaces where semantic search is enabled, it also finds pages that are close in meaning (a question in English about a page in French, a paraphrase…). These results are labeled “by meaning”, or “words and meaning”. Enabling semantic search on a space never pushes down its pages found by words.

Notes dropped but not yet tidied that match are shown above the results, under Waiting to be tidied: what was just dropped can be found right away.

Search covers:

Markdown and all text files;
the text of PDFs, including scanned ones thanks to OCR;
the text of Word, Excel, PowerPoint and OpenDocument documents;
the text of images (OCR);
file names.

The text of a new PDF becomes searchable a few seconds to a few minutes after it is added, the time to extract it. You never see a result from a space you do not have access to.

Inbox and activity

Inbox (left bar): filter by text or by type (notes, changes, items left by the tidy-up). On a note, a contributor can edit it (if the tidy-up had left it, it goes back to “to process”), move it to the inbox of another space where they contribute, or delete it (it stays in the Git history). A changed page that asks nothing of the tidy-up can be marked as seen.
Activity: everything that changes in the space, according to the Git history (pages, inbox, tidy-ups), with commits pushed outside Synaptane flagged. Each page touched leads to the detail of the change.