Confide Content — Jung Rama

Overview

Enterprise CMSs are static repositories. The documents just sit there.

Companies store thousands of contracts, specs, and manuals in systems that can only find them by name. The knowledge inside stays locked up, and the processes around those documents — extracting fields, comparing vendors, translating, summarizing — stay manual. Confide Content combines CMS storage, retrieval-augmented generation, and low-code workflow orchestration in one platform.

It's built as three independent modules — CMS, Search + AI, and Workflow — so different clients enable different combinations, behind enterprise auth (SSO/SAML) and role-based access control.

Confide Content file manager — folders of documents with owners, modification history, and per-file actions including Chat with AI — The content hub

The journey

One document, from upload to an answer with sources.

01
Documents come in
CMS
Teams upload and organize content — vendor PDFs, spreadsheets, images — with versioning, metadata tagging, and access control from the first file.
02
Files become viewable and readable
Conversion
An async pipeline converts PDFs and spreadsheets into browser-viewable formats and extracts the text the AI layer will work with.
03
Content gets indexed
Retrieval
Documents are chunked and embedded into a per-tenant index — Japanese included — so every file becomes searchable by meaning, not just filename.
04
Answers come with receipts
Search + AI
Chat and search run over the index. AI outputs carry confidence scores and citations back to the source documents.
05
Processes run themselves
Workflow
Drag OCR, translation, and summarization blocks into a graph, point it at a folder, and the pipeline runs in the background — resumable if a step fails.

Inside the platform

What working with it actually looks like.

Pages you write, files you read — same place

Content isn't only uploads. A block-based editor lets teams author pages right inside the CMS, next to the files they reference — and PDFs open in an in-browser viewer, converted by the async pipeline, so nobody downloads a file just to read it.

Block-based page editor with headings, lists, and rich text blocks — Block-based page editor

In-browser PDF viewer rendering a technical product catalog — In-browser document viewer

Ask the folder, not a person

Any file or folder can become a conversation. The chat panel answers from the documents themselves — scoped to what the user is allowed to see, with the sources pinned right above the answer. Ask about a product catalog and it reads the actual catalogs, then tells you what's in them, document by document.

Ask AI panel answering questions about a folder of product catalogs, with sources attached — RAG chat with cited sources

Pipelines you drag together

The workflow builder is a node graph: model nodes (object detection, OCR, classification, content moderation), transforms (crop, text extraction, frame sampling), and logic (confidence filters, tagging) — wired from inputs to outputs and run against folders of content.

Node-graph workflow builder wiring an image input through a content moderation model and blur visualization to a safe-image output — Low-code workflow builder

Enterprise guardrails by default

Every page keeps an activity log and a version history you can restore from. Sharing is restricted-by-default, granted per person or group with explicit roles — the access model the retrieval layer enforces when the AI answers.

Info panel showing activity and version history with restore — Version history & activity

Share dialog with restricted access, member roles, and link copying — Access control

Three modules

Storage, intelligence, and automation — enabled per client.

CMS

File management

Storage with versioning, metadata tagging, org/user/permission management, and object casting — turn a file into a task, or content into a PDF.

Search + AI

Chat and search

Retrieval-augmented chat and hybrid search over everything a user is allowed to see, with citations linking every answer to its sources.

Workflow

Automated pipelines

A low-code node graph for document processes — extract fields from vendor PDFs, generate comparison matrices, chain OCR to translation to summary.

Architecture

A content core, a retrieval service, and an action layer.

The CMS core owns storage, versioning, metadata, and access control. Retrieval is its own service behind a hardened gateway: the gateway terminates TLS and validates tokens, while the retrieval API itself lives on a private network with separate stores for job metadata and the chunk index. Above both sits the action layer — the CMS talks to workflows only through versioned action contracts, and every execution writes to an authoritative trace and audit history.

CMS core

Storage, versioning, metadata tagging, org/user/ACL management, async conversion pipeline.

Retrieval service

Gateway-only exposure, per-tenant chunk index, hybrid dense + BM25 search fused with RRF.

Action layer

Versioned action contracts between CMS and workflows, with trace IDs and audit events on every run.

Outcome

Documents that answer back — safely.

The shift is in how people reach information: instead of opening files one by one and skimming, they ask — and the answer arrives with the documents that back it up.

Every answer: Cited to its source; AI responses pin their sources above the answer, so trust doesn't depend on faith in the model — you can click through and check.
4 nodes: Instead of a codebase; Document processes that used to mean custom scripts — moderation, OCR, tagging — become a small graph anyone can read, wire, and rerun.
Per-tenant: Isolation by default; Access control reaches inside the AI: retrieval only surfaces what each user is allowed to see, and each client's data stays in its own index.

Next project

SaaS product

Galleryou

Events · 2026

A CMS that reads its own documents.