One Helm chart. Your cluster. Your data.
The same five stateless services that run our SaaS install on EKS, GKE, AKS, OpenShift, or on-premises.
Data stores stay outside the application pods, every pod has a hard memory limit, and there is no broker to run.
Why it does not fall over.
The lesson from bundled replay stacks.
All-in-one stacks crash from memory for predictable reasons: several memory-hungry stateful systems share one undersized node with the application, ingest buffers without bounds when a downstream is slow, and pods have no limits so the kernel kills a random victim.
Backlog lives at the edge: the browser's bounded buffer backs off; the gateway never grows.
Every pod is Guaranteed QoS, with the Go memory limit set from the container limit.
Object storage is the durable log. No Kafka, no Kinesis, nothing else stateful in the cluster.
Stateful stores run with their own hard limits, on their own node pool in the bundled profile.
A load test on every merge slows the object store on purpose and fails if gateway memory rises.
helm repo add backstory https://charts.backstory.io
helm install backstory backstory/backstory \
--namespace backstory --create-namespace \
--set profile=external \
--set datastores.postgres.url="postgres://…" \
--set datastores.clickhouse.addr="clickhouse:9000" \
--set datastores.s3.endpoint="https://s3.internal" \
--set datastores.s3.bucket="backstory-chunks" Bring your own, or use operators.
| Profile | Who it is for | What the chart installs | Minimum sizing |
|---|---|---|---|
| external (recommended) | Enterprises with a platform team | Our five services and add-ons. You supply Postgres, ClickHouse, S3-compatible storage, and for live view either a Valkey you run or ElastiCache Serverless, plus a voice relay if you want voice. | 2 nodes × 4 vCPU / 8 GB for our services |
| bundled | Evaluations, small deployments, air-gapped sites | Everything above plus CloudNativePG, the ClickHouse operator, MinIO, Valkey, and the voice relay with TURN | 1 node × 8 vCPU / 32 GB, or 2 × 4 vCPU / 16 GB; 500 GB SSD |
| compose | Laptop evaluation | Same images with Postgres, ClickHouse, and MinIO in Docker | 16 GB laptop |
The bundled profile refuses to install on a node smaller than 16 GB rather than letting it crash later. A sizing calculator prints recommended volume sizes from your expected sessions and retention.
What runs where.
- gateway ingest, acks after storage
- worker process, signals, AI, retention
- api REST, auth, live
- ai-gateway AI routing, your keys
- web static dashboard
Same release train
SaaS and self-hosted customers get the same images on the same day. Upgrades are a Helm upgrade with expand-and-contract migrations.
Configurable retention
Per-project retention from 1 to 400 days, enforced by object lifecycle rules keyed on a retention class tag, identical on S3 and MinIO.
Licensing
Self-hosted Enterprise uses the same meters as SaaS. Usage is reported by the chart's metering job or a signed monthly report for air-gapped sites.
Product updates by email.
Leave a work email and we will tell you when something ships. Or book a demo with an engineer.
You are on the list.
Product updates will go to .
Free for 1,000 stored sessions a month, with unlimited seats. Create a workspace and add one script tag.
Start