Skip to main content
Migrating between deployment targets requires only two things: a valid API key for the destination, and an updated endpoint in your SDK configuration. Because replication is write-behind and endpoint-agnostic, the same encrypted snapshot payloads that target your self-hosted instance can be redirected to the managed Platform (or back) with no schema changes.

Self-Hosted → Managed Platform

1

Register at app.sovseal.com

Go to app.sovseal.com and create an account. If your team already has one, ask an admin to invite you.
2

Generate a Platform API key

From the Devices page, create a new key. It will have the prefix sov_live_. Copy it immediately — it is shown once.
3

Update your SDK configuration

Replace the self-hosted endpoint with the managed one:
4

Replicate existing records

There is no automatic migration of previously-replicated data between endpoints — each endpoint only ever holds what was written to it. Your local LanceDB store is unaffected either way; going forward, new writes replicate to whichever endpoint the client points at.To carry forward history that only exists on the old endpoint, walk it with restore() / lineage() against the old client and re-snapshot() it against the new one — the same primitives documented in Verified Semantic Recall.

Managed Platform → Self-Hosted

1

Deploy your self-hosted instance

Follow Deploy Self-Hosted. Confirm the instance is reachable with the verification call shown there before proceeding.
2

Update the endpoint

3

Records already on the Platform stay there

Switching endpoints does not delete or move anything. Records replicated to the Platform remain in the Platform’s database. If you need them removed, that is a request to security@sovseal.com — there is no self-service bulk-delete today.

Why this is safe either way

Your local LanceDB database is always the source of truth for reads. Recall latency and accuracy never depend on which replication endpoint is configured — a wrong or unreachable endpoint degrades sync durability, not agent behavior. Switching back is the same one-line change in reverse.