Self-hosting
Basemind runs anywhere Docker runs. Choose the setup that fits your team.
Option 1: single-user / small team (recommended to start)
docker-compose.single-user.yml runs one backend worker, one combined Celery
worker with the scheduler, PostgreSQL, Redis and Caddy. It idles around
1 GB of RAM.
git clone <your fork> basemind && cd basemind
cp .env.single-user.example .env.single-user
# edit .env.single-user: SECRET_KEY, DATABASE_PASSWORD, REDIS_PASSWORD,
# BASEROW_PUBLIC_URL
docker compose -f docker-compose.single-user.yml --env-file .env.single-user up -d --build
Open http://localhost:4000 (or your BASEROW_PUBLIC_URL) and create the
first account; it becomes the instance administrator.
Option 2: full Docker Compose
docker-compose.yml runs separate services (backend, web-frontend, Celery
workers for normal and export jobs, beat, Caddy, PostgreSQL, Redis) and scales
to many users. Copy .env.example to .env, set the secrets and run
docker compose up -d.
Running on a public domain with HTTPS
Point the domain’s DNS A record to the server, open ports 80 and 443, and
set in .env.single-user:
BASEROW_PUBLIC_URL=https://basemind.example.com
BASEROW_CADDY_ADDRESSES=https://basemind.example.com
WEB_FRONTEND_PORT=80
WEB_FRONTEND_SSL_PORT=443
HOST_PUBLISH_IP=0.0.0.0
Caddy obtains and renews a Let’s Encrypt certificate automatically.
For published applications on subdomains, also set
BASEROW_BUILDER_DOMAINS=example.com, add a wildcard DNS record
*.example.com and include https://*.example.com in
BASEROW_CADDY_ADDRESSES.
Small servers
On a server with 1–2 GB of RAM:
- add a 2 GB swap file;
- set
CELERY_POOL=soloand a smallPOSTGRES_SHARED_BUFFERS(e.g.64MB); - build the images on another machine (a front-end build needs about 4 GB)
and copy them over with
docker save | ssh ... docker load.
Important environment variables
| Variable | Purpose |
|---|---|
BASEROW_PUBLIC_URL |
The URL people use to open Basemind. |
SECRET_KEY, BASEROW_JWT_SIGNING_KEY |
Secrets for signing. Keep them stable and private. |
DATABASE_HOST/NAME/USER/PASSWORD |
PostgreSQL connection. |
REDIS_URL or REDIS_HOST/PASSWORD |
Redis connection. |
EMAIL_SMTP, EMAIL_SMTP_HOST, EMAIL_SMTP_PORT, EMAIL_SMTP_USER, EMAIL_SMTP_PASSWORD, FROM_EMAIL |
Outgoing email (invitations, password resets, notifications). |
BASEROW_BUILDER_DOMAINS |
Domains for published applications. |
BASEROW_EMBEDDINGS_API_URL |
Embeddings service for semantic memory search and the assistant knowledge base. |
BASEROW_ENTERPRISE_ASSISTANT_LLM_MODEL |
Fallback model for the in-app assistant. |
HOURS_UNTIL_TRASH_PERMANENTLY_DELETED |
How long trashed items can be restored. |
The full list is in docs/installation/configuration.md in the repository.
Semantic memory search
Run the embeddings service (the embeddings/ folder of the repository) and set
BASEROW_EMBEDDINGS_API_URL to its address. Then rebuild existing vectors:
docker compose exec backend ./baserow reindex_project_memories
Backups
Everything lives in PostgreSQL and the media volume:
docker compose exec db pg_dump -U baserow -Fc baserow > basemind.dump
docker run --rm -v <project>_media:/media -v $PWD:/backup alpine \
tar czf /backup/media.tgz -C /media .
Never point two stacks at the same PostgreSQL volume.
Upgrading
git pull
docker compose -f docker-compose.single-user.yml --env-file .env.single-user up -d --build
Migrations run automatically when the backend starts and are designed to run without downtime.