Database | VexAI Docs

SQLite (Default)

The default backend requires zero configuration and is ideal for single-server deployments.

WAL mode: enables concurrent read access without blocking
FTS5 full-text search: fast text indexing on messages
Embedding storage: separate message_embeddings table for AI vectors
Database file: data/vexai.db

💡 Tip

SQLite is production-ready for single-server bots. You only need PostgreSQL when handling multiple shards or very high traffic.

PostgreSQL

For production environments with high concurrency and advanced vector search.

Connection pooling: configurable min/max connections
tsvector full-text search: native PostgreSQL text indexing
pgvector: native vector column type with IVFFlat indexing for embeddings
Easy setup: use docker-compose.postgres.yml

Configuration (JSON)

{
  "database": {
    "type": "postgres",
    "url": "postgres://vexai:password@localhost:5432/vexai",
    "pool": { "min": 2, "max": 10 }
  }
}

Environment Variables

DATABASE_TYPE=postgres
DATABASE_URL=postgres://vexai:password@localhost:5432/vexai

Comparison

Feature	SQLite	PostgreSQL
Setup	Zero-config	Requires server
Full-text search	FTS5	tsvector
Embedding storage	Separate table	pgvector native
Indexing	B-tree	B-tree + IVFFlat
Concurrency	WAL (single-writer)	Full MVCC
Connection pooling	N/A	Built-in
Best for	Development, single server	Production, high traffic

Live Migration

Seamlessly migrate between backends without downtime.

/migrate to-postgres: streams all data from SQLite to PostgreSQL with a live progress bar
/migrate status: shows the current backend and row counts per table

⚠️ Warning

Back up your SQLite database before migrating. Copy data/vexai.db to a safe location first.

Database Tables

Sequential migrations (v1–v8) create the following tables:

Table	Purpose
`logs`	Structured bot activity logs
`cron_jobs`	Scheduled task definitions
`users`	User records
`warnings`	User warning history
`reminders`	Time-based reminders
`auto_roles`	Roles assigned on join
`welcome_config`	Welcome message settings
`model_overrides`	Per-user LLM model overrides
`conversations` / `conversation_messages`	Persistent conversation history
`semantic_memories`	FTS5-powered memory store
`messages` / `message_revisions`	Discord message mirror + audit trail
`message_embeddings`	AI embedding vectors

Embedding Providers

VexAI supports multiple embedding providers for generating semantic vectors.

Provider	Default Model	Dimensions	Notes
`builtin`	`Xenova/all-MiniLM-L6-v2`	384	In-process ONNX, offline, zero cost
`openai`	`text-embedding-3-small`	1536	Best quality
`openrouter`	`openai/text-embedding-3-small`	1536	Many models, one key
`local`	`nomic-embed-text`	768	Any OpenAI-compatible endpoint

Configuration

{
  "embeddings": {
    "enabled": true,
    "provider": "builtin",
    "model": "Xenova/all-MiniLM-L6-v2",
    "dimensions": 384,
    "batchSize": 50
  }
}

Semantic Search

AI embedding-powered meaning-based search that finds related messages even without exact keyword matches.

Converts messages into high-dimensional vectors capturing their meaning
Finds semantically similar content across the entire message history
Background worker generates embeddings for all messages automatically
Supports similarity thresholds and result limits for precise querying