Better than Supabase: agentic infrastructure is vastly more than Postgres

Supabase earned its place. Postgres with auth, storage, functions and realtime in one dashboard is a great way to build a web app, and Postgres is the right default database. Paddington runs on it.

Agents change what an app needs from its backend. A web app answers requests. An agent works: it runs for minutes or hours, calls models and tools, waits on people, retries, remembers, and acts on systems that matter. Here is what that takes, and where it comes from on Kilter.

An agent needs	Kilter catalog
Durable work that survives restarts	temporal, dapr
Events and queues between agents	nats, kafka, redpanda, rabbitmq
Memory beyond a chat log	honcho, letta, qdrant, falkordb
One gateway for every model call	bifrost
Identity for agents as well as people	ory, zitadel, keycloak, authentik
Fine-grained permissions	spicedb
Secrets the agent never sees	openbao
Policy the platform enforces	kyverno, Airbender ARC
Search over what the business knows	meilisearch, typesense, elasticsearch
Traces of what the agent did	signoz, otel-collector, tempo, loki
Business apps agents can operate	twenty, outline, nextcloud, ghost
Long-running compute in any language	Kubernetes, which Kilter runs for you

Supabase answers one row of that table very well. Kilter answers all of them with one file:

# kilter.yaml
services:
  - postgres
  - ory
  - temporal
  - nats
  - bifrost
  - qdrant
  - signoz

It is lighter than it sounds

A full Kilter dev stack runs in under 3 GB (20 pods, 2.0 GB measured 3 Nov); Supabase's own Docker setup takes 2 to 4 GB for 12 containers. You get the platform for about the cost of the database.

And it stays yours

The same kilter.yaml runs in any Paddington region, on a Miniscaler or on your own cluster. Moving is a deploy.

Use Supabase for a web app with a database. Use Kilter when agents do the work.