Skip to content

Commit cae3c91

Browse files
PhilipinhoPhilipinho
committed
AI search docs
1 parent 3cf8926 commit cae3c91

5 files changed

Lines changed: 222 additions & 1 deletion

File tree

‎docs/self-hosting/ai/_category_.json‎

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
{
2+
"label": "AI",
3+
"position": 4,
4+
"collapsed": false
5+
}

‎docs/self-hosting/ai/ai-search/ai-search.md‎

Lines changed: 180 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,180 @@
1+
---
2+
tags:
3+
- Enterprise
4+
---
5+
6+
# AI Search Setup
7+
8+
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.
9+
10+
## Prerequisites
11+
12+
1. **PostgreSQL with pgvector extension** - Required for storing vector embeddings
13+
2. **AI provider account** - OpenAI, Google Gemini, or self-hosted Ollama
14+
15+
## Installing pgvector
16+
17+
### Docker Setup (Recommended)
18+
19+
Use the official pgvector Docker image instead of the standard PostgreSQL image:
20+
21+
```yaml
22+
# docker-compose.yml
23+
services:
24+
db:
25+
image: pgvector/pgvector:pg17
26+
environment:
27+
POSTGRES_DB: docmost
28+
POSTGRES_USER: docmost
29+
POSTGRES_PASSWORD: your_password
30+
volumes:
31+
- db_data:/var/lib/postgresql/data
32+
```
33+
34+
The pgvector extension will be automatically available.
35+
36+
### Manual Installation
37+
If you are using a non-docker installation of Postgres, you can manually install the pgvector extension.
38+
See the pgvector installation guide: https://github.com/pgvector/pgvector?tab=readme-ov-file#installation
39+
40+
## Supported AI Providers
41+
42+
Docmost supports three AI providers: OpenAI (Azure OpenAI), Google Gemini, and Ollama (local LLMs).
43+
44+
### Provider Configuration
45+
46+
All providers require these base environment variables:
47+
48+
```bash
49+
AI_DRIVER=<provider> # openai, gemini, or ollama
50+
AI_EMBEDDING_MODEL=<model> # Model used for generating embeddings
51+
AI_COMPLETION_MODEL=<model> # Model used for answering questions
52+
```
53+
54+
**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.
55+
56+
---
57+
58+
## OpenAI Configuration
59+
60+
Supports OpenAI API and Azure OpenAI.
61+
62+
### Environment Variables
63+
64+
```bash
65+
AI_DRIVER=openai
66+
OPENAI_API_KEY=sk-proj-xxxxx
67+
AI_EMBEDDING_MODEL=text-embedding-3-small
68+
AI_COMPLETION_MODEL=gpt-4o-mini
69+
```
70+
71+
**Optional:**
72+
```bash
73+
OPENAI_API_URL=https://api.openai.com/v1 # For Azure or custom endpoints
74+
```
75+
76+
### OpenAI Preset Models
77+
78+
| Model |
79+
|-------|
80+
| `text-embedding-3-small` |
81+
| `text-embedding-3-large` |
82+
| `text-embedding-ada-002` |
83+
84+
### Example
85+
86+
```bash
87+
AI_DRIVER=openai
88+
OPENAI_API_KEY=sk-proj-xxxxxxxxxxxxx
89+
AI_EMBEDDING_MODEL=text-embedding-3-small
90+
AI_COMPLETION_MODEL=gpt-4o-mini
91+
```
92+
93+
---
94+
95+
## Google Gemini Configuration
96+
97+
### Environment Variables
98+
99+
```bash
100+
AI_DRIVER=gemini
101+
GEMINI_API_KEY=AIzaSyxxxxx
102+
AI_EMBEDDING_MODEL=gemini-embedding-001
103+
AI_COMPLETION_MODEL=gemini-2.5-flash
104+
```
105+
106+
### Gemini Preset Models
107+
108+
| Model |
109+
|-------|
110+
| `gemini-embedding-001` |
111+
112+
---
113+
114+
## Ollama Configuration
115+
Docmost AI search and embeddings supports local LLMs via Ollama.
116+
Ollama Docker Installation Guide: https://docs.ollama.com/docker
117+
118+
### Environment Variables
119+
120+
```bash
121+
AI_DRIVER=ollama
122+
OLLAMA_API_URL=http://localhost:11434
123+
AI_EMBEDDING_MODEL=nomic-embed-text
124+
AI_COMPLETION_MODEL=qwen2.5:7b
125+
```
126+
127+
### Ollama Preset Models
128+
129+
| Model |
130+
|-------|
131+
| `nomic-embed-text` |
132+
| `qwen3-embedding` |
133+
134+
### Setup Ollama
135+
136+
1. Install Ollama: https://ollama.com/download
137+
2. Pull embedding model:
138+
```bash
139+
ollama pull nomic-embed-text
140+
```
141+
3. Pull completion model:
142+
```bash
143+
ollama pull qwen2.5:7b
144+
```
145+
146+
### Example
147+
148+
```bash
149+
AI_DRIVER=ollama
150+
OLLAMA_API_URL=http://localhost:11434
151+
AI_EMBEDDING_MODEL=nomic-embed-text
152+
AI_COMPLETION_MODEL=qwen2.5:7b
153+
AI_EMBEDDING_DIMENSION=768
154+
```
155+
156+
For Docker deployments, use the Ollama container:
157+
```yaml
158+
services:
159+
ollama:
160+
image: ollama/ollama:latest
161+
ports:
162+
- "11434:11434"
163+
volumes:
164+
- ollama_data:/root/.ollama
165+
```
166+
167+
Then set `OLLAMA_API_URL=http://ollama:11434` in your Docmost container.
168+
169+
---
170+
171+
## Enable AI Search in Workspace
172+
173+
After configuring your AI provider:
174+
175+
1. Log in to Docmost as a workspace admin
176+
2. Go to **Settings** → **AI settings**
177+
3. Toggle **AI-powered search (Ask AI)** to enable
178+
4. Wait for background job to generate embeddings for existing pages (monitor via logs)
179+
180+
**Note:** Embeddings are generated asynchronously. New pages get embeddings on creation/update. Existing pages are queued for processing when AI search is enabled.

‎docs/self-hosting/configuration/index.md‎

Lines changed: 27 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -62,6 +62,33 @@ The maximum size limit for an import file.
6262
FILE_IMPORT_SIZE_LIMIT=200mb
6363
```
6464

65+
## AI
66+
### OpenAI / Azure OpenAI
67+
```shell
68+
AI_DRIVER=openai
69+
OPENAI_API_KEY=sk-proj-xxxxxxxxxxxxx
70+
AI_EMBEDDING_MODEL=text-embedding-3-small
71+
AI_COMPLETION_MODEL=gpt-4o-mini
72+
# optional
73+
OPENAI_API_URL=
74+
```
75+
76+
### Google Gemini
77+
```shell
78+
AI_DRIVER=gemini
79+
GEMINI_API_KEY=AIzaSyxxxxx
80+
AI_EMBEDDING_MODEL=gemini-embedding-001
81+
AI_COMPLETION_MODEL=gemini-2.5-flash
82+
```
83+
84+
### Ollama
85+
```shell
86+
AI_DRIVER=ollama
87+
OLLAMA_API_URL=http://localhost:11434
88+
AI_EMBEDDING_MODEL=nomic-embed-text
89+
AI_COMPLETION_MODEL=qwen2.5:7b
90+
```
91+
6592
## Telemetry
6693
We anonymously collect the active version, user count, page count, space and workspace count.
6794
To disable telemetry:

‎docs/self-hosting/development.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
---
2-
sidebar_position: 4
2+
sidebar_position: 5
33
---
44

55
# Development

‎docs/self-hosting/environment-variables.md‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -86,6 +86,15 @@ The default Draw.io embed url is `https://embed.diagrams.net`.
8686
|-----------| ------------------------ |----------------------------|
8787
| `DRAWIO_URL` | `https://embed.diagrams.net` | Your custom draw.io server url. |
8888

89+
## Search Configuration (Enterprise)
90+
91+
| Variable | Example | Description |
92+
|---------------------|---------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
93+
| `SEARCH_DRIVER` | `typesense` | The search engine driver to use. Default is `postgres` (database search). `typesense` is a purpose-built search engine with typo-tolerant, fast search available in Enterprise edition. |
94+
| `TYPESENSE_URL` | `http://localhost:8108` | The URL of your Typesense server. Required when `SEARCH_DRIVER=typesense`. (Enterprise) |
95+
| `TYPESENSE_API_KEY` | `your_api_key_here` | Your Typesense API key with read/write permissions. Required when `SEARCH_DRIVER=typesense`. (Enterprise) |
96+
| `TYPESENSE_LOCALE` | `en` | The locale for text analysis and search. Default is `en`. Examples: `en`, `es`, `fr`, `de`, `ja`, `zh`, `ko`, etc. See [Typesense supported languages](https://typesense.org/docs/latest/api/search.html#supported-languages) for full list. (Enterprise) |
97+
8998
## Telemetry
9099
We anonymously collect the active version, user count, page count, space and workspace count.
91100

0 commit comments

Comments
 (0)