Skip to main content

AI Search Setup

AI-powered search in Docmost uses vector embeddings to provide semantic search across your workspace. This is an enterprise feature that requires a valid license key.

Prerequisites​

  1. PostgreSQL with pgvector extension - Required for storing vector embeddings
  2. AI provider account - OpenAI, Google Gemini, or self-hosted Ollama

Installing pgvector​

Docker Setup​

warning

Since Docmost's default PostgreSQL compose image is Alpine-based, and pgvector does not have an Alpine build, you will not be able to do a clean in-place swap. Instead, you'll need to migrate your data.

You can switch your PostgreSQL image to pgvector's bundled PostgreSQL build.

If you already have Docmost set up, you may want to back up your PostgreSQL database first, then import it into the pgvector PostgreSQL image.

See pgvector docker guide: https://github.com/pgvector/pgvector#docker

Manual Installation​

If you are using a non-docker installation of Postgres, you can manually install the pgvector extension. See the pgvector installation guide: https://github.com/pgvector/pgvector#installation

Supported AI Providers​

Docmost supports three AI providers: OpenAI (Azure OpenAI), Google Gemini, and Ollama (local LLMs).

Provider Configuration​

All providers require these base environment variables:

AI_DRIVER=<provider>           # openai, gemini, or ollama
AI_EMBEDDING_MODEL=<model>     # Model used for generating embeddings
AI_COMPLETION_MODEL=<model>    # Model used for answering questions

Important: AI_EMBEDDING_DIMENSION is optional and auto-detected from preset models. Only set it manually if using a custom model not in the preset list.

OpenAI Configuration​

Supports OpenAI API and Azure OpenAI.

Environment Variables​

AI_DRIVER=openai
OPENAI_API_KEY=sk-proj-xxxxx
AI_EMBEDDING_MODEL=text-embedding-3-small
AI_COMPLETION_MODEL=gpt-4o-mini

Optional:

OPENAI_API_URL=https://api.openai.com/v1  # For Azure or custom endpoints

OpenAI Preset Models​

Model
text-embedding-3-small
text-embedding-3-large
text-embedding-ada-002

Example​

AI_DRIVER=openai
OPENAI_API_KEY=sk-proj-xxxxxxxxxxxxx
AI_EMBEDDING_MODEL=text-embedding-3-small
AI_COMPLETION_MODEL=gpt-4o-mini

Google Gemini Configuration​

Environment Variables​

AI_DRIVER=gemini
GEMINI_API_KEY=AIzaSyxxxxx
AI_EMBEDDING_MODEL=gemini-embedding-001
AI_COMPLETION_MODEL=gemini-2.5-flash

Gemini Preset Models​

Model
gemini-embedding-001

Ollama Configuration​

Docmost AI search and embeddings supports local LLMs via Ollama.
Ollama Docker Installation Guide: https://docs.ollama.com/docker

Environment Variables​

AI_DRIVER=ollama
OLLAMA_API_URL=http://localhost:11434
AI_EMBEDDING_MODEL=nomic-embed-text
AI_COMPLETION_MODEL=qwen2.5:7b

Ollama Preset Models​

Model
nomic-embed-text
qwen3-embedding

Setup Ollama​

  1. Install Ollama: https://ollama.com/download
  2. Pull embedding model: 
    ollama pull nomic-embed-text
  3. Pull completion model: 
    ollama pull qwen2.5:7b

Example​

AI_DRIVER=ollama
OLLAMA_API_URL=http://localhost:11434
AI_EMBEDDING_MODEL=nomic-embed-text
AI_COMPLETION_MODEL=qwen2.5:7b
AI_EMBEDDING_DIMENSION=768

For Docker deployments, use the Ollama container:

services:
  ollama:
    image: ollama/ollama:latest
    ports:
      - "11434:11434"
    volumes:
      - ollama_data:/root/.ollama

Then set OLLAMA_API_URL=http://ollama:11434 in your Docmost container.

Enable AI Search in Workspace​

After configuring your AI provider:

  1. Log in to Docmost as a workspace admin
  2. Go to Settings → AI settings
  3. Toggle AI-powered search (Ask AI) to enable
  4. Wait for background job to generate embeddings for existing pages (monitor via logs)

Note: Embeddings are generated asynchronously. New pages get embeddings on creation/update. Existing pages are queued for processing when AI search is enabled.