Data services

Rivtor runs its own Postgres, Redis and storage on the same private network as your apps, so a query never leaves the data centre and app-to-database latency stays under 1 ms.

Data services are created when detection finds code that needs them, or when a service lists them under data in deploy.toml. Provisioning runs in parallel with the build and takes under 2 seconds. Their credentials are stored encrypted and injected as environment variables at runtime.

Injected variables

environment of a project called my-app

# Postgres
DATABASE_URL=postgresql://p_my_app:<password>@host.docker.internal:6432/p_my_app
DIRECT_DATABASE_URL=postgresql://p_my_app:<password>@host.docker.internal:6432/p_my_app
PGHOST=host.docker.internal
PGPORT=6432
PGDATABASE=p_my_app
PGUSER=p_my_app
PGPASSWORD=<password>

# Redis
REDIS_URL=redis://default:<password>@redis:6379

# R2 object storage (AWS_* aliases are set too)
S3_ENDPOINT=https://<account>.r2.cloudflarestorage.com
S3_BUCKET=rv-my-app
S3_REGION=auto
S3_ACCESS_KEY_ID=<key>
S3_SECRET_ACCESS_KEY=<secret>

Postgres 17

Each project gets its own database p_<slug> and its own role with the same name and a random 32-character password. Other roles cannot connect to it.
Apps connect through PgBouncer in transaction mode on port 6432, so serverless-style apps such as Next.js do not exhaust connections. Each role is limited to 20 connections and a 60-second statement timeout.
DIRECT_DATABASE_URL is provided for tools such as Prisma that ask for one; in v1 it is the same as DATABASE_URL.
There is no public database endpoint. Use the db tool for queries, temporary connections and exports.
Over the plan's size limit, the database becomes read-only until you free space or upgrade. Nothing is deleted.
Databases never sleep, on any plan.

Redis 7

Each project gets its own Redis on the project's private network, reachable at redis:6379 with a password.
Memory is capped at the plan's Redis size with the allkeys-lru policy: when full, the least recently used keys are evicted. Use it as a cache and queue, not as the only copy of your data.

Object storage (R2)

Each project gets a Cloudflare R2 bucket rv-<slug> and an API key scoped to that bucket only.
It speaks the S3 API: point any S3 SDK at S3_ENDPOINT with region auto. The same values are also set as AWS_* variables for SDKs that read those.
No egress fees, and files can be served through Cloudflare's CDN.

Sizes per plan

Data service sizes per plan
	Free	Hobby	Pro	Team
Postgres	1 GB per project	5 GB total	20 GB total	80 GB total
Redis	50 MB	100 MB	256 MB	1 GB
File storage	1 GB	10 GB	50 GB	200 GB
Point-in-time backups	7 days	14 days	30 days	30 days
Database branches	No	No	Yes	Yes

Backups

Postgres is archived continuously (write-ahead log) with nightly base backups, so a database can be restored to any point in the backup window of its plan: 7 days on Free, 14 on Hobby, 30 on Pro and Team. Every week a random database is restored automatically to prove the backups work.

Export

Your data is yours. Download a standard dump at any time:

$npx rivtor db export -o my-app.dump

$pg_restore --no-owner -d postgres://localhost/my_app my-app.dump

Read-only queries from your terminal or agent: npx rivtor db query "select …" or the db tool. They run in a read-only transaction with a 5-second timeout and return up to 500 rows.