Everything an app backend needs. Nothing to wire up.
A database you design at runtime, an API that enforces your rules, auth that can be the identity
provider for your other apps, files and realtime. One process, one data directory.
Data
Access rules
One expression per operation, compiled to safe, parameterized SQL. Rules are checked live in the admin UI as you type. New collections are locked by default.
postsrule
listpublished = true || owner = @request.auth.id
viewpublished = true || owner = @request.auth.id
create@request.body.owner = @request.auth.id
updateowner = @request.auth.id
delete locked; superusers only
list rule → SQLillustrative
SELECT … FROM "posts"WHERE "published" = ?1 OR "owner" = ?2-- ?1 = 1, ?2 = <id from the auth token>
Data
Collections & REST API
Define tables at runtime in the admin UI and get a full REST API: filtering, sorting, pagination, relation expansion and batch transactions. Real SQLite columns, not JSON blobs.
rackend is an OIDC provider itself. Your apps log in through it with any standard OIDC library, on a hosted, themeable login page. No separate identity server to run.
your app
rackend login page
id_token + access_token
issuer
https://your-rackend.example
flow
authorization code + PKCE
scopes
openid profile email offline_access
Auth
Authentication
Email and password, passkeys, one-time codes and magic links. Plus login with any OIDC provider: paste the issuer URL and rackend discovers the rest.
Google
Microsoft
GitHub
Authentik
Pocket ID
Keycloak
any OIDC issuer
Realtime
Realtime
Subscribe to record changes over SSE or WebSocket. Every event is checked against access rules, per subscriber. Ephemeral channels for presence and typing indicators.
Local disk by default, or any S3-compatible store. Access is controlled by the same rules as records.
local disk
AWS S3
Cloudflare R2
Garage
RustFS
Operations
Observability
An audit trail of every change: who, what, when, before and after. Request logs, app and host metrics, a built-in dashboard, Prometheus and OpenTelemetry export.
Images & thumbnails
EXIF and GPS metadata stripped on upload, orientation fixed, thumbnails generated on demand as WebP.
Collaborative editing
Yjs-compatible collaborative documents out of the box. Works with Tiptap, BlockNote and CodeMirror.
OpenAPI for every instance
An OpenAPI document generated from your live schema, typed per collection. Generate a client in any language.
Standard errors
Every API error is an RFC 9457 problem details object with a stable, machine-readable code.
Admin UI
Embedded in the binary: schema editor, records browser, auth and provider setup, dashboard, settings.
03/Data safety
Built to keep your data.
Data safety wins over everything else in rackend's design. Upgrading should be the most boring
thing you do all week.
Upgrades never lose data
A verified backup before every schema migration, integrity checks after. If anything looks wrong, like an accidental downgrade, rackend refuses to start rather than risk your data.
Every upgrade path tested
In CI, the database of every released version is upgraded to the latest one and checked.
A trash, not a drop
Deleted fields and collections are kept in a trash for 30 days before they are purged.
Backups are a file copy
One SQLite file plus a storage folder. Scheduled backups are built in.
One writer, always
An exclusive lock stops a second instance from writing to the same data directory.
upgrade · v1.3 → v1.4illustrative
$ docker compose pull && docker compose up -d
✓ data directory locked rackend.lock✓ migration checksums verified
✓ backup written backups/pre-upgrade-12-to-14-….db✓ backup integrity_check ok✓ applied migration 13
✓ applied migration 14
✓ integrity_check, foreign_key_check ok→ serving on 0.0.0.0:8080
accidental downgradeexit 78
✗ database schema v14 is newer than this binary supports (v12).
Downgrading is not supported; restore a pre-upgrade backup
from /data/backups to go back.
04/Operations
One process. One data directory.
No Postgres, Redis or queue to run next to it. One static binary or one small Docker image,
for amd64 and arm64.
Zero config to start
Everything is configured at runtime in the admin UI. Environment variables exist only for bootstrapping: listen address, data directory, an optional encryption key.
services: rackend: image: ghcr.io/axanet/rackend:1.0 # pin a minor version ports: ["8080:8080"] volumes: ["rackend_data:/data"]volumes: rackend_data:
Runs where you already are
a €5 VPS
a Raspberry Pi-class board
your homelab
Coolify
Unraid
Inside the binary
A hexagonal core with no I/O of its own. Everything it talks to sits behind a port, so new adapters such as gRPC or MCP can be added later without touching it.
rackend · one process
Adapters in
HTTPREST · SSE · WebSocket · OIDC
gRPClater
MCPlater
Core
collections
rules
auth · OIDC
files
realtime
audit
Adapters out
SQLite
filesdisk or S3
images
mail
Data directory
data.db
logs.db
storage/
backups/
05/How rackend differs
Standing on PocketBase's shoulders.
rackend is inspired by PocketBase, and shares its single-binary
spirit. Here is where it takes a different path, and what it deliberately doesn't do.
vs. PocketBase
Same idea, more built in.
A built-in OIDC provider, so other apps can log in through rackend
Collaborative editing with Yjs-compatible documents
An audit trail and app + host metrics out of the box
Written in Rust
vs. Supabase / Appwrite
No stack to operate.
One process instead of a multi-service deployment
SQLite instead of Postgres
Aimed at single-app, single-server deployments
Honest trade-offs
What it won't do.
Single instance; no horizontal scaling
No custom server-side code (hooks) in v1
One app per instance; no multi-tenancy
Own your backend.
One command away. Free and open source.
$docker run -p 8080:8080 -v rackend_data:/data ghcr.io/axanet/rackend:0.0