Skip to Content
EnterpriseDeployment Options

Deployment Options

Omniflow runs in four shapes. Pick based on your compliance posture, data residency requirements, and operations capacity.

The four options

OptionWhere it runsBest for
Full SaaS (multi-tenant)Omniflow’s managed cloudMost customers — fastest to deploy, lowest operational overhead.
Private cloud (single-tenant)Dedicated AWS / Azure / GCP project, region of your choiceRegulated industries, custom retention, custom DLP, mTLS.
Exoscale (Switzerland)Exoscale’s Swiss datacentersFINMA-regulated workloads, Swiss data residency.
On-premiseCustomer’s own infrastructureAir-gapped or sovereign-cloud requirements.

Multi-tenant SaaS — isolation model

In SaaS, every customer is a workspace in a multi-tenant database. Isolation is enforced at the row level using PostgreSQL Row-Level Security (RLS):

  • Every business table has a tenant_id column.
  • Every RLS policy gates reads and writes on tenant_id = auth.tenant().
  • Service-role connections (edge functions) explicitly scope queries by tenant_id.
  • Cross-tenant access is blocked by default at the database level — bugs in application code can’t expose other customers’ data.
PropertySaaS
DatabaseShared PostgreSQL cluster, RLS-enforced row isolation.
Object storagePer-tenant prefixed bucket paths; ACLs scoped per tenant.
Vector indexPer-agent index, scoped by tenant.
Voice runtimeShared Railway runtime, JWT-scoped per tenant.
Background workersShared, with per-tenant rate limits.

For customers with stricter isolation requirements, private cloud moves the database, storage, and runtime into a dedicated environment.

Private cloud — what’s dedicated

ComponentDedicated
DatabaseDedicated PostgreSQL instance in your region.
Object storageDedicated bucket.
Voice runtimeDedicated Railway / Fly project.
Edge functionsDedicated Supabase project.
CDN / ingressDedicated (regional).

Source code is provided to customers in private cloud deployments under the customer-source license — read access for security review, no redistribution.

Exoscale (Switzerland)

For Swiss data residency and FINMA-regulated workloads:

  • Deployed in Exoscale’s Geneva (DK-GVA-2) and Zurich (CH-ZH-1) datacenters.
  • ISO 27001 / SOC 2 inherited from Exoscale infrastructure.
  • Support staff with Swiss residency.
  • All customer data — conversations, transcripts, embeddings, audio — stays in Switzerland.
  • LLM providers configurable: managed routing to EU/Swiss inference endpoints, or BYO model running inside Exoscale.

On-premise

For air-gapped or sovereign-cloud deployments:

ComponentReference spec (per 100 concurrent voice + 1k concurrent chat)
Application servers4× 16-core, 64 GB RAM.
Database1× primary, 1× replica; 16-core, 128 GB RAM, NVMe SSD.
Object storageS3-compatible (MinIO / Ceph), 5 TB minimum.
GPU (if BYO inference)2× A100 80 GB or equivalent for self-hosted LLM.
Network10 Gbps internal; egress sized to expected channel volume.

A reference Helm chart and Terraform module are provided. On-premise deployments include source code, an annual upgrade support contract, and a dedicated solutions architect.

On-premise customers run their own monitoring, backups, and patching. Omniflow provides recommended dashboards and runbooks.

Datacenter locations (managed deployments)

RegionDatacenter
EU-CentralAWS Frankfurt
EU-SwitzerlandExoscale Geneva, Exoscale Zurich
US-EastAWS Virginia
US-WestAWS Oregon
APAC-SouthAWS Singapore
APAC-SoutheastAWS Sydney

Production monitoring

Omniflow runs:

  • Metrics: Prometheus + Grafana for system metrics; Datadog for APM.
  • Logs: Centralized logging with retention per the workspace policy.
  • Tracing: OpenTelemetry across edge functions, voice runtime, and DB.
  • Alerting: PagerDuty for engineering on-call; per-customer Slack/email digests for customer-facing incidents.
  • Synthetic checks: Continuous synthetic conversations across all regions; alarms fire on degradation.
  • Status page: https://status.omniflow.example mirrors real-time service health.

For private cloud and on-prem, the same dashboards and alert templates are provided as code (Terraform).

Roadmap (next 24 months)

The product roadmap is shared with active customers under NDA. High-level public posture:

  • 0–6 months: Deeper self-service automation library, expanded agent reasoning models, Polish and Bahasa GA.
  • 6–12 months: Multi-modal retrieval (image + diagram), expanded BYO-LLM support, enhanced Speech Analytics dashboards.
  • 12–24 months: Advanced agent orchestration (multi-agent), cross-tenant federation for franchise/agency use cases, Voice Bot fine-tuning UI.

Customers on private-cloud and Exoscale tiers receive a quarterly roadmap briefing under NDA.

Open in Omniflow

If you want to…Go to
Read security detailsSecurity & Compliance
Configure data retentionSettings overview
Read the runtime architectureVoice Runtime