Dedicated cluster
When to choose a dedicated AACsearch cluster, what it includes, and how it differs from self-hosted.
Dedicated cluster
A dedicated cluster is an AACsearch deployment whose search nodes and database are reserved for a single customer. It is run by us, in the region you choose, on infrastructure that no other customer shares.
This is not self-hosted. The cluster lives in our cloud account, with our operations team on call. Self-hosted (you run the cluster on your own infrastructure) is on the roadmap — see the bottom of this page.
When you need it
Pick a dedicated cluster if any of these apply:
- Your compliance program prohibits sharing infrastructure with other customers, even with logical isolation.
- Your search workload is bursty enough that the shared cluster's per-tenant rate limits and quotas are a real constraint, not a theoretical one.
- You need a region that is not on our standard list (EU / US / RU) — for example, India, Australia, Singapore, or a specific Russian region for 152-FZ purposes.
- You need to pin a specific Typesense or PostgreSQL version, e.g. to match an external integration.
If your reason is "we are concerned about the multi-tenant security model", read the Tenant isolation page first. The shared cluster's isolation guarantee is the same as a dedicated cluster's — what's different on dedicated is noisy-neighbor performance, not data isolation.
What's included
| Component | Shared cluster | Dedicated cluster |
|---|---|---|
| Search nodes | Shared, autoscaled | Reserved, sized to your spec |
| PostgreSQL | Shared with row-level isolation | Reserved instance |
| Object storage | Shared bucket, key-prefixed | Reserved bucket |
| Region | EU / US / RU | Any AWS / GCP region we operate in (custom) |
| Network | Internet-facing | Optionally VPC-peered or via private endpoint |
| Backups | Cluster-wide WAL-G + S3 snapshots | Independent, your retention policy |
| Maintenance windows | We set them | Negotiated, off your business hours |
| Typesense version | Latest stable | Pinned to a version, upgrade on your schedule |
| Custom SLA | 99.9 % shared | Up to 99.99 %, negotiated |
The application surface (oRPC API, dashboard, widget, connectors, SDKs) is identical — same code, same endpoints. Your application code does not need to change to move from shared to dedicated.
How sizing works
We size the cluster around three numbers:
- Peak concurrent QPS — the busiest second of your busiest day.
- Index size — total document count and average document size, across all collections.
- Reindex window — the longest you can tolerate an alias swap to take.
Underspeccing makes search slow during peaks. Overspeccing burns money. We deliberately size at p99 + 30 % headroom and revisit every quarter. You get an invoice line we can defend, not an opaque number.
How procurement works
- Scoping call. 30 minutes to confirm the use case and the constraints.
- Sizing. We send a written sizing proposal with QPS and index size assumptions. Push back on anything that doesn't match your reality.
- Contract. MSA + DPA + the cluster appendix (region, SLA, support tier).
- Provisioning. 5–10 business days once the contract is signed.
- Data migration. If you're on a shared cluster today, we run a migration to copy your indexes. Zero downtime via alias swap.
- Go-live. Switch DNS / dashboard target. Old shared-cluster keys are revoked.
We will not promise a faster timeline than what we can defend operationally. If a region requires a new control plane to be stood up, we say so.
What is not included (yet)
- Self-hosted / air-gapped. Running the cluster inside your own VPC, on your own AWS / GCP account, or in a fully offline environment, is roadmap. The current dedicated cluster is in our cloud, period. See "Self-hosted" below.
- White-label / OEM. Re-skinning the dashboard with your own brand for resale is roadmap. We can do a custom domain (e.g.
search.yourbrand.com) on the cluster API today; we cannot rebrand the dashboard UI yet. - Custom modifications to the search engine. We run upstream Typesense with our extensions. Per-customer forks are not on the roadmap.
Self-hosted
Status: roadmap. A self-hosted AACsearch distribution that you run inside your own VPC or on-prem is not generally available. The current "private deployment" option is a dedicated cluster in our cloud, with network controls (VPC peering, private endpoint) on top.
If self-hosted is a hard requirement (e.g. air-gapped government deployment), please email sales@aacsearch.com so we can sequence the work. We will not bill for or commit to a delivery date for self-hosted until we publish a release here.
SLA
The standard shared-cluster SLA is 99.9 % uptime per calendar month, measured against search query availability, with a credit schedule for breaches. Dedicated clusters can negotiate:
- 99.95 % or 99.99 % targets, depending on regional infrastructure and replication topology.
- Custom credit schedules tied to your contract value.
- A P1 response time of 15 minutes, 24/7, instead of business-hours best-effort.
The number is meaningful only if both sides understand how it's measured. We use synthetic search probes from at least three geographic regions per cluster region. Maintenance windows are excluded only if pre-announced ≥ 7 days in advance.