One product, two places to run it today, a third planned

• Shared PostgreSQL with mandatory workspace scope, application authorization and row-level security.
• Planned: a dedicated database, storage and worker deployment for contracts that need stronger operational isolation.
• Each workspace policy sets the allowed processing regions, providers and models, and one retention period.
• Rate limits per route and a voice-call cap per workspace. Fair scheduling between tenants is planned.

Teams that want to start fast and accept a qualified cloud region.

• API and console, PostgreSQL, a Redis-compatible broker, S3-compatible object storage, workers, and nginx as the single published origin.
• Local speech to text runs on CPU when you install the local AI option. GPU workers and speaker separation are planned. Neo4j is optional.
• No application container has a route to the internet. Outbound traffic, if any, passes through a deny-by-default proxy with your allowlist.
• Sign-in with e-mail and password, or Google when your network allows it. Keys live in your own secrets storage, as files or variables.
• Health and readiness endpoints, migration job, drain and rollback steps documented in the runbook.

Organizations whose recordings must never leave their network.

• Audio, transcripts, identity, search and evidence stay local. The installation keeps working offline.
• Planned: an optional management plane receives only approved health, version and license counters, and you can inspect the outgoing payload.
• A second mode allows local speech to text and masking, then cloud reasoning that gets masked text by default, opt-in per workspace.
• A failed local model never silently falls back to a cloud API.

Organizations that want private data with selected cloud capacity.