Why Atlas Features How it works REST API FAQ Impact Inspector Controller MergeSense Contact Show interest
Organization wiki & knowledge base

Your organization's knowledge base, built to power your AI pipelines too.

PrashFlow Atlas is a self-hosted wiki and knowledge base for organizations: folders, versioned pages and uploaded documents, permissions down to a single document, and watermarked viewing when it matters. Every page and file is also reachable through an authenticated API, so your own AI and RAG pipelines can integrate directly with your knowledge base instead of it staying trapped in a browser tab.

An integration can never read more than the person who authorized it can see in the app. Same permission engine, browser or pipeline.

Knowledge base sync, in progress
AUTH Your pipeline authenticates, resolved to its owner's own permissions
SCAN Every wiki and document that owner can read, organized and labeled
SYNC 47 documents handed to your pipeline, ready to ingest
1
API to integrate your entire knowledge base with an AI or RAG pipeline, no scraping required
6
permission actions per folder or document: view, fence-view, download, upload, edit, and more
0
extra services required to get clean documents and metadata out, your pipeline pulls them directly
100%
self-hosted on infrastructure you own, storage and all
Why Atlas

A wiki that was built to be read by more than a browser

Most internal wikis stop at rendering a page in a tab. PrashFlow Atlas keeps that, but treats every page and file as a permissioned, versioned, auditable object that your own tools can reach too, through the same access rules a person would hit clicking around.

Pages you type, files you upload, one model

Write a page straight in the browser, in Markdown, with images dropped in, or upload a file. Either way it's a versioned document in a folder, same permission and audit rules either path.

Permissions down to a single document

A folder-inheriting engine: grant a group or a person view, download, upload, or edit rights on a whole folder, then narrow it for one document without touching the parent grant.

Watermarked when it matters

PDFs and Office documents render with the viewer's name, email, IP, and timestamp burned server-side into every page, not a client-side overlay someone can inspect away.

Built to be read by your own tools

A token-authenticated REST API lists and downloads exactly what its owner can see in the app. Point a RAG ingestion job at it and let it decide what changed since the last run.

An audit trail that doesn't lie

Every login, view, download, and permission change is written to an append-only log, no update or delete path exists in the code, exportable as CSV for a real review.

Own your infrastructure, own your data

Self-hosted on infrastructure you control, the same architecture as every other PrashFlow product.

The problem

Most wikis were never built to be queried, only browsed

They work fine until you need to feed one to anything other than a person.

Your knowledge base is trapped in a browser tab

No API means a RAG pipeline or any other tool has to scrape rendered HTML, or someone exports pages by hand. Either way, nothing stays current.

Access control is all-or-nothing

A wiki that's either fully open internally or locked behind one blanket permission leaves no room for a folder that only one team, or one external contractor, should see.

No record of who actually read what

When a sensitive page leaks, or an external reader downloads something they shouldn't have, there's usually no log to check, and no mark on the file to trace it back.

How it works

Create, grant, read, query

The same four stages, whether the reader is a person or a pipeline.

1

Create

Type a page in Markdown, drop in images, or upload a file. Folders, versioning, and password protection all built in.

2

Grant

Give a group or an individual person exact actions on an exact folder or document. Nearest grant wins, nothing merges upward.

3

Read

Pages render in-browser, PDFs and Office files watermarked with the reader's identity burned in. A normal login session, nothing extra.

4

Query

An API token authenticates external tools against the exact same permission rules, for a RAG pipeline or anything else that needs programmatic access.

Everything included

What's actually built today

No roadmap items, no "coming soon." This is what runs in the app and API right now.

Wikis, folders, pages, and documents

Multiple wikis per organization, nested folders, typed Markdown pages with inline images, uploaded files with full version history: list, restore, or delete a specific version, and a trash with configurable auto-purge.

Groups and invitations

Invite external readers into a group, scoped to the folders that group was actually granted, with optional domain and IP allowlisting.

NDA gate

An optional click-through NDA a reader has to accept before their first view, recorded with a timestamp and IP address.

Token-authenticated REST API

Personal access tokens, scoped to read or export, list every wiki and document a token's owner can see, download a document's original bytes, or pull a bulk manifest built for a RAG ingestion job: name, mime type, folder path, size, version, and a ready-to-call download URL for each document.

Append-only audit log

Every login, view, download, upload, and permission change recorded, exportable as CSV, with no update or delete path anywhere in the service layer.

Password-protected documents

Strong encryption at rest, keyed to a password only the uploader sets, not stored anywhere: lose it, and that document is unreadable, by design, even for an administrator.

Reader engagement scoring

Weighted activity per reader in a wiki, views and downloads ranked, so an admin can see who's actually reading it, not just who has access.

REST API

Built to feed a RAG pipeline, not just a browser tab.

An API token authenticates as whichever user created it, so every call is filtered through that person's own permission grants, a token can never read more than they could see in the app. The export endpoint returns a manifest with everything a RAG ingestion job needs to decide what changed and pull it, in one call.

  • List every wiki and every document a token can see, folder path and mime type included
  • One bulk export call returns the full manifest, name, path, size, version, and a downloadUrl per document
  • Scoped tokens, read or export, revocable any time from the dashboard, never a shared API key
LIST Every document in a wiki, folder path included
FOUND "Runbooks/Deploy.md", ready to pull
FETCH Original file bytes, decrypted and ready to ingest
SCOPE Read or export, checked on every call
Watermarked viewing

A screenshot still exists. It's not anonymous.

No web page can reliably block an OS-level screenshot, so PrashFlow Atlas doesn't pretend to. Every PDF and Office document is rendered server-side, per request, with the viewer's name, email, IP address, and timestamp burned directly into the page. A capture still identifies whoever took it. Screenshot-attempt heuristics log the attempt too, on top of that.

  • Rendered server-side per request, not a CSS overlay a reader can disable
  • Text, position, opacity, rotation, and which fields to burn in, configurable per organization
  • A distinct "fence view" tier for restricted, view-only access, tracked separately in the audit log
Deploy Runbook.pdf, watermarked view
VIEW jane.doe@partnerco.demo · 203.0.113.5
LOG DOCUMENT_VIEWED, fenceView: false
TILE CONFIDENTIAL, burned in at 16% opacity, every page
Security by design

Every request checked, browser or API token

There's exactly one permission engine. A session and an API token both resolve to a normal authenticated user, and that user's grants are what every read, download, or list call is filtered through, no separate, looser path for automation.

Browser session or API token
resolves to one authenticated user
Permission engine
folder-inheriting, nearest grant wins
Your wiki content
never more than that user was granted
API tokens carry the creating user's own permissions, never more Password-protected documents strongly encrypted at rest Append-only audit log, no update or delete path exposed Watermark burned server-side, not a client-side overlay Screenshot-attempt heuristic logged to the audit trail
Self-hosted, on your terms

Your wiki. Your data. Your control.

We deploy the full stack directly on infrastructure you own. Nothing about your documents, your pages, or who read what ever leaves your environment.

FAQ

Questions, answered

Yes, that's what the REST API's export endpoint is for: a manifest of every document a token can read, with metadata and a downloadUrl each. Embedding, chunking, and the vector store are intentionally left to your own pipeline, not built into Atlas.
No. PDFs and Office documents (Word, Excel, PowerPoint) render watermarked. Everything else, images, video, audio, Markdown pages, code, previews inline unwatermarked. Screenshot-attempt logging still applies regardless of file type.
Both. Type a page straight in the browser in Markdown, with images dropped in, or upload a file. Either way it becomes a versioned document with the same permission and audit rules.
Not today, there's no channel for the password in a stateless API call, so those documents return 403 from the REST API and need to go through the web app instead.
Self-hosted only today, deployed entirely on infrastructure you own and control. A managed/hosted offering isn't built yet.

Give your wiki an API, not just a browser tab.

Self-hosted, permission-checked either way, and ready to feed a RAG pipeline today.

Questions first? contact@prashflow.com