Compendiq

Open-source AI knowledge base for Confluence Data Center
Self-hosted. On-premise. Your data never leaves your network.

Install · Features · Architecture · Docs · Community · Security

---

Compendiq connects to your Confluence Data Center instance, syncs your pages, and gives you AI superpowers: ask questions across your entire knowledge base, improve articles with one click, generate new documentation from templates, and detect knowledge gaps -- all running on hardware you control.

Why this exists: Confluence Data Center has no native AI features. Cloud-only AI tools don't work for on-premise deployments. Compendiq fills that gap without sending a single byte to external servers (when using Ollama for local inference).

Install in 3 Minutes

From zero to setup wizard with a single command:

curl -fsSL https://raw.githubusercontent.com/Compendiq/compendiq-ce/main/scripts/install.sh | bash

Requirements: Docker Engine 24+ with Compose v2, 4 GB RAM, port 8081 free.

The installer generates cryptographic secrets, pulls images from GHCR, starts the stack (frontend, backend, PostgreSQL, Redis), and opens the setup wizard in your browser. Ollama must be running on the host for local AI inference.

What the installer does (step by step)

1. Generates JWT_SECRET and PAT_ENCRYPTION_KEY (AES-256, 32+ chars)
2. Writes ~/compendiq/docker-compose.yml with all secrets embedded
3. Pulls images from ghcr.io/compendiq/compendiq-ce-*
4. Starts 4 containers: frontend (nginx), backend (Fastify), PostgreSQL 17 (pgvector), Redis 8
5. Waits for the health endpoint (up to 3 min)
6. Opens http://localhost:8081 -- the first-run wizard

Custom install directory / Ollama URL

# Custom install location
INSTALL_DIR=~/mydir curl -fsSL https://raw.githubusercontent.com/Compendiq/compendiq-ce/main/scripts/install.sh | bash

# Remote Ollama instance
OLLAMA_BASE_URL=http://my-gpu-server:11434 curl -fsSL ... | bash

Uninstall

bash ~/compendiq/uninstall.sh

Stops containers, removes volumes, deletes the install directory.

Pull the models

Compendiq needs an embedding model and a chat model:

ollama pull bge-m3          # Required: embeddings (1024 dimensions)
ollama pull qwen3:4b        # Or any chat model you prefer

---

What It Does

AI-Powered Q&A Across Your Knowledge Base

Ask questions in natural language and get answers sourced from your Confluence pages. Powered by RAG with pgvector hybrid search (vector cosine similarity + full-text keyword search + Reciprocal Rank Fusion).

One-Click Article Improvement

Select an article and improve it for grammar, structure, clarity, technical accuracy, or completeness. Generate new articles from templates (runbook, how-to, architecture, troubleshooting). Summarize long documents. Auto-tag pages with LLM classification.

Full Confluence Data Center Integration

Bidirectional sync with XHTML storage format conversion. Round-trip support for Confluence macros: code blocks, task lists, panels, user mentions, page links, draw.io diagrams (read-only rendering), Children Pages (expandable inline), and Attachments with download links.

More Features

Category	What you get
Editor	TipTap v3 with Vim mode, drag-and-drop blocks, find-and-replace, image/table captions, header numbering, code block language detection, clipboard image paste
AI	Multi-provider LLM (Ollama default, or any OpenAI-compatible API), real-time SSE streaming, conversation history, content summarization, knowledge gap detection, duplicate page detection
Security	AES-256-GCM PAT encryption, JWT with refresh token rotation, RBAC with custom roles, OIDC/SSO (Enterprise), rate limiting, SSRF protection, prompt injection guard
Analytics	Page views, engagement metrics, search pattern tracking, knowledge graph visualization
Operations	PDF import/export, page verification workflow, knowledge requests, audit logging, OpenTelemetry tracing

---

Architecture

Confluence Data Center (XHTML Storage Format)
    |  REST API v1 (Bearer PAT)
    v
Backend (Fastify 5 + TypeScript + Node.js 22)
    |-- Sync Service ............ polls Confluence, stores in PostgreSQL
    |-- Content Converter ....... XHTML <-> HTML <-> Markdown
    |-- Embedding Service ....... chunks text, generates vectors via Ollama
    |-- RAG Service ............. hybrid search (pgvector + FTS + RRF)
    |-- Redis Cache ............. hot cache with TTL eviction
    v
Frontend (React 19 + Vite + TailwindCSS 4)
    |-- TipTap v3 Editor (HTML round-trip with Confluence macros)
    |-- AI Assistant (SSE streaming)
    |-- Glassmorphic UI (Radix UI + Framer Motion)

Tech Stack

Layer	Technology
Backend	Fastify 5, TypeScript, Node.js 22+
Frontend	React 19, Vite 7, TailwindCSS 4, Radix UI, Zustand, TanStack Query, Framer Motion
Editor	TipTap v3 (ProseMirror)
Database	PostgreSQL 17 + pgvector
Cache	Redis 8
AI/ML	Ollama (local) or OpenAI-compatible APIs, bge-m3 embeddings (1024 dims)
Auth	JWT (jose) + bcrypt + optional OIDC/SSO
Validation	Zod schemas shared via @compendiq/contracts
Testing	Vitest, Playwright, @testing-library/react
Infra	Docker Compose (4 services), multi-stage Dockerfiles, GHCR

Data Flow

Confluence pages are synced over REST, converted from XHTML to HTML and Markdown, chunked and embedded for vector search, then served through a React frontend with a real-time AI chat interface. Every LLM call streams via Server-Sent Events. All data stays in your PostgreSQL database on your hardware.

---

Community Edition vs Enterprise

Compendiq is open-core. The Community Edition (this repo) is fully functional under AGPL-3.0 with no artificial limits.

Feature	Community (free)	Enterprise (paid)
Confluence sync + AI Q&A + article generation	Yes	Yes
RAG search (vector + keyword + hybrid)	Yes	Yes
Multi-provider LLM (Ollama, OpenAI, Azure, vLLM)	Yes	Yes
Rich text editor with Confluence macros	Yes	Yes
PDF import/export	Yes	Yes
RBAC with custom roles	Yes	Yes
Audit logging	Yes	Yes
OIDC/SSO	--	Yes
Per-space RAG permissions	--	Yes
Audit log export	--	Yes
Priority support	--	Yes

Interested in Enterprise? Open an issue or reach out via GitHub Discussions.

---

Developer Quick Start

Full local development setup

1. Clone and install

git clone https://github.com/Compendiq/compendiq-ce.git
cd compendiq-ce
npm install

2. Configure environment

cp .env.example .env
# Edit .env -- set JWT_SECRET and PAT_ENCRYPTION_KEY to random 32+ char strings

3. Start infrastructure

docker compose -f docker/docker-compose.yml up -d   # PostgreSQL + Redis

4. Start dev servers

npm run dev   # Backend (3051) + Frontend (5273) with hot reload

5. Create your account

Open http://localhost:5273. First user gets admin role. Configure Confluence URL and PAT in Settings.

Running tests

npm test                          # All tests (contracts + backend + frontend)
npm run test -w backend           # Backend only (needs PostgreSQL on port 5433)
npm run test -w frontend          # Frontend only (jsdom)
npm run test:e2e                  # E2E (Playwright, needs running servers)
npm run lint                      # ESLint
npm run typecheck                 # TypeScript strict mode

---

Configuration

All configuration is via environment variables. Key settings:

Variable	Required	Description
JWT_SECRET	Yes	JWT signing secret (32+ characters)
PAT_ENCRYPTION_KEY	Yes	AES-256-GCM key for Confluence PATs (32+ characters)
OLLAMA_BASE_URL	--	Ollama server URL (default: http://localhost:11434)
EMBEDDING_MODEL	--	Embedding model (default: bge-m3)
LLM_PROVIDER	--	ollama (default) or openai
OPENAI_BASE_URL	--	OpenAI-compatible API URL (for Azure, LM Studio, vLLM, etc.)
OPENAI_API_KEY	--	API key when using OpenAI provider
POSTGRES_URL	--	PostgreSQL connection string
REDIS_URL	--	Redis connection string

Full configuration reference

Variable	Default	Description
JWT_SECRET	--	JWT signing secret (32+ chars, must not be default in production)
PAT_ENCRYPTION_KEY	--	AES-256-GCM key for encrypting Confluence PATs (32+ chars)
POSTGRES_USER	kb_user	PostgreSQL username
POSTGRES_PASSWORD	changeme-postgres	PostgreSQL password
POSTGRES_DB	kb_creator	PostgreSQL database name
POSTGRES_URL	postgresql://kb_user:changeme-postgres@localhost:5432/kb_creator	Full connection string
REDIS_PASSWORD	changeme-redis	Redis password
REDIS_URL	redis://:changeme-redis@localhost:6379	Full Redis connection string
OLLAMA_BASE_URL	http://localhost:11434	Ollama server URL
EMBEDDING_MODEL	bge-m3	Server-wide embedding model (1024 dimensions)
LLM_PROVIDER	ollama	LLM provider: ollama or openai
LLM_BEARER_TOKEN	--	Bearer token for authenticated Ollama/LLM proxies
LLM_VERIFY_SSL	true	Set to false for self-signed LLM certs
LLM_STREAM_TIMEOUT_MS	300000	Streaming request timeout (ms)
LLM_CACHE_TTL	3600	Redis TTL for LLM cache (seconds)
OPENAI_BASE_URL	--	OpenAI-compatible API base URL
OPENAI_API_KEY	--	API key (required with openai provider)
DEFAULT_LLM_MODEL	--	Fallback model for background workers
SYNC_INTERVAL_MIN	15	Confluence sync polling interval (minutes)
CONFLUENCE_VERIFY_SSL	true	Set to false for self-signed Confluence certs
NODE_ENV	development	development or production
OTEL_ENABLED	false	Enable OpenTelemetry tracing
OTEL_EXPORTER_OTLP_ENDPOINT	--	OTLP collector endpoint

---

API

Interactive API documentation is available at /api/docs (Swagger UI) when the backend is running.

Endpoint	Description
GET /api/health	Health checks (live, ready, start probes)
POST /api/auth/*	Register, login, refresh, logout
GET/POST/PUT/DELETE /api/pages/*	Page CRUD, versions, tags, embeddings, duplicates
POST /api/llm/*	AI operations (improve, generate, summarize, ask, chat)
POST /api/sync	Manual Confluence sync
GET /api/search	Keyword, semantic, and hybrid search
GET/POST /api/admin/*	Admin settings, audit log, RBAC, LLM config

All endpoints except /api/health and /api/auth/* require a JWT Bearer token.

---

Contributing

Contributions are welcome. See CONTRIBUTING.md for the full guide.

Quick version: branch from dev as feature/<description>, add tests, run npm test && npm run lint && npm run typecheck, open a PR to dev.

Security

Found a vulnerability? Report it via GitHub Security Advisories (private channel). Do not open a public issue. See SECURITY.md for details.

License

GNU Affero General Public License v3.0

---

Built by a solo developer in Europe. If Compendiq saves your team time, star the repo -- it helps others find it.