RMT Engineering Logo
Hosting, residency and uptime

Deployment & Hosting

OptiML ships as one managed cloud platform. Name your region and the tenant is up in minutes, with RMT operating it from there. On Enterprise, the data itself can live in your own cloud account under keys you hold.

  • Provisioned in your chosen region
  • 231 tables under row-level security
  • Encryption keys you can revoke
We reply within one business day. No newsletter.
99.9% Monthly uptime commitment
231 Tables under row-level security
16+ Regional compliance packs
Your keys Held in your own KMS

One product, one release train, and the region is your call

There is no menu of topologies here. There is one product. It runs in OptiML Cloud, in the region you choose, on the same release train as every other tenant, which is why an upgrade never turns into a migration project. What the Enterprise plan adds is where your data rests, and who holds the keys to it.

We run it. You choose where the data lives.

The interesting decisions on a hosting page are usually made for you somewhere in a roadmap you never see, and then handed back to you as flexibility, which is a generous word for a dropdown with two options in it. Here the split runs the other way. We own the continuous, unglamorous work of keeping the platform up: the patching, the on-call rota, the restore drill nobody wants to schedule. What is left for you is the part that actually turns up in your obligations. Which region your tenant sits in. How long each class of data is kept. And on Enterprise, whose cloud account the database lives in.

How it fits together

The path a conversation takes through Deployment & Hosting

Under customer-managed data residency the PostgreSQL database and the S3-compatible recording and object storage run in the customer's own cloud account or datacentre, while the application, AI runtime, media plane and analytics store keep running in OptiML Cloud, connected privately and mutually authenticated YOUR CLOUD ACCOUNT OR DATACENTRE PostgreSQL database THE ONLY DURABLE COPY S3-compatible recording and object storage THE ONLY DURABLE COPY ENCRYPTION KEYS HELD BY YOU, IN YOUR KMS PRIVATE, MUTUALLY AUTHENTICATED PROCESSED IN TRANSIT · NO PRIMARY COPY OPTIML CLOUD · YOUR CHOSEN REGION Application, desk and console AI runtime and model routing Media plane Analytics and reporting store PERSISTS WITH RMT: PII-REDACTED OPERATIONAL TELEMETRY AND A BOUNDED MESSAGE-QUEUE WINDOW BOTH ENUMERATED BY DATA CLASS IN THE DPA
By the numbers

Few Facts - at a glance

  • 99.9%

    monthly uptime commitment in the SLA, with a published credit schedule

  • 231

    tables under PostgreSQL row-level security, so isolation is enforced by the database

  • 16+

    regional compliance packs — GDPR, HIPAA, CCPA, TCPA, PCI-DSS, DPDP, FINRA, the EU AI Act and more

The process

How It Works

The four stages, end to end

Getting onto the platform is a provisioning conversation rather than an implementation project. Two of the four stages below are yours: naming the region, and writing the configuration. Two are ours. The diagram underneath covers the one Enterprise option that changes this picture, customer-managed data residency, where the durable copy of your data sits in your own cloud account while the rest of the platform keeps running in OptiML Cloud.

  1. You pick the region

    You name the region your tenant is provisioned in. From then on that choice governs where conversation data, transcripts, recordings and analytics get stored. Retention is set at the same time, one window per class of data. After that it enforces itself. Nobody has to remember to run a cleanup job.

  2. The tenant comes up

    Provisioning takes minutes, not a quarter. The tenant arrives inside OptiML Cloud with row-level security already applied across 231 tables and tenant context derived from the authenticated token. The compliance pack for your jurisdiction is switched on before anybody logs in.

  3. You configure it as code

    Agents, flows, knowledge sources, phone numbers, integrations, users and roles all live in the Terraform provider, so a change to any of them is a merge request rather than a support ticket someone at RMT picks up when they get to it. Somebody on your side reviews it. The REST API, the CLI and signed webhooks cover whatever the provider does not.

  4. We keep it running

    Patching, monitoring, autoscaling, backups, verified restores and security operations run on our side, against the 99.9% monthly commitment. You still get to watch. OpenTelemetry traces, Prometheus metrics and PII-redacted logs are exported into your own monitoring stack, so the platform is not a black box with a status light on the front of it.

Features

Every capability you need in one module

1. One platform, run by us

A tenant is a provisioning step, not a project. From there the region you picked governs where everything the platform records ends up, and retention enforces itself.

OptiML is a managed cloud platform. Your tenant is provisioned in minutes and hosted in the region you choose, so data stays where you need it. Isolation is not a code convention. PostgreSQL row-level security on 231 tables means the database refuses to return another tenant's row, whatever the application asks for. RMT operates the platform against a 99.9% monthly uptime commitment: patching, monitoring, scaling, backup and disaster-recovery drills. Enterprise buyers can take a dedicated single-tenant instance, still inside OptiML Cloud, still fully managed, with its own retention windows and compliance posture.

  • Provisioned in minutes, in the region you choose
  • Patching, monitoring, scaling, backup and DR drills, all run by RMT, so you are not staffing an ops rota to cover three in the morning
  • 99.9% monthly uptime, with credits
  • Dedicated single-tenant hosting on Enterprise

See the governance, guardrail and audit layer

2. Your data stays in the region you choose

Conversation data, transcripts, recordings and analytics are stored in the region you nominated when the tenant was provisioned, and they do not wander from it. Retention is configurable per tenant and per class of data, and the platform enforces it on its own. Regional compliance packs, sixteen and counting, carry the jurisdiction-specific rules for consent, disclosure, retention and erasure. Opening in a new market is configuration. It is not a project.

GDPR HIPAA CCPA TCPA PCI-DSS DPDP FINRA EU AI Act and more

3. Isolation the database enforces, not the code

This is the difference between a control you can audit and a convention you have to trust. A defect in application code cannot leak data across tenants here, because the application is not what is enforcing the boundary.

Most multi-tenant platforms enforce isolation in application code: a WHERE tenant_id = ? clause on every query, correct right up until the day one query gets written without it. OptiML enforces it in PostgreSQL instead, with row-level security policies on 231 tables. The database refuses to return another tenant's row regardless of what the application asks for. Tenant context is derived from the authenticated token and fails closed. No resolvable tenant, no data. For a BPO the consequence is commercial. Each client programme is an isolated tenant carrying its own retention, compliance mode, audit trail and DNC list, so an audit can be answered one programme at a time instead of across the whole estate.

  • Row-level security policies on 231 tables
  • Tenant context derived from the authenticated token, so a missing or forged tenant id gets no rows back at all
  • Fails closed
  • Retention, compliance mode, audit trail and DNC list, all set per client programme

How per-programme DNC lists work in Outbound & Dialer

4. A dedicated instance, still fully managed

Two Enterprise options sit here, and they answer different questions. One isolates the instance. The other moves the system of record into your cloud account, under your keys.

Dedicated single-tenant hosting is an Enterprise option: an isolated instance inside OptiML Cloud, in your chosen region, with its own retention windows, its own compliance posture and no shared infrastructure underneath it. RMT still runs it. Same release train, same monitoring, same on-call rota. Nothing about the product changes. What changes is that no other tenant is sharing the machinery.

5. Your data, your account

Some obligations are about where the data lives, not where the software runs. Customer-managed data residency is an option on the Enterprise plan. It puts your database and recording storage in your own cloud account or your own datacentre, reached over a private, mutually authenticated connection. You hold the encryption keys, so access is yours to grant and yours to revoke. The system of record is yours: those two stores hold the only durable copy of your conversations, transcripts and recordings. The platform itself keeps running in OptiML Cloud. Application, AI runtime, media plane and console process that data in transit and keep no primary copy of it. What persists with us is PII-redacted operational telemetry and a bounded message-queue retention window, both enumerated by data class in the DPA, so your reviewer gets a list to read line by line and does not have to take it on trust from whoever is on the call.

The platform itself runs in OptiML Cloud

This option moves your data, not the software. Worth being blunt about it, because buyers often read "your own cloud account" as a private installation of the whole product and then discover otherwise during implementation. The application, AI runtime, media plane and console keep running in OptiML Cloud on the same release train as everyone else. The DPA enumerates by data class exactly what is left behind with RMT.

Request the reference architecture under NDA

Customer-managed data residency — mechanism
What runs in your accountPostgreSQL database and S3-compatible recording and object storage
What runs in OptiML CloudApplication, AI runtime, media plane, Agent Desk, Supervisor Console, admin
What transits OptiML CloudLive audio and media, transcripts and prompts in process, and model-provider requests — processed, not stored as a primary copy
What persists in OptiML CloudPII-redacted operational logs and traces, and a bounded message-queue retention window; enumerated by data class in the DPA
Analytics and reporting storeStays in OptiML Cloud, in your chosen region, under the same regional and retention controls; reporting rows carry PII redaction rather than relocation
ConnectionPrivate connectivity between the platform and your data plane, mutually authenticated
KeysCustomer-held; envelope encryption with per-recording keys wrapped by your tenant key in your KMS
Supported clouds and connectivityNamed per engagement. The reference architecture states the cloud or datacentre in scope and the private-connectivity mechanism for it
What you provideThe cloud account or datacentre, a network path for the private connection, a KMS to hold the tenant key, and the operational commitments in the shared-responsibility row
AvailabilityEnterprise plan, scoped and priced per engagement. Reference architecture and shared-responsibility matrix available under NDA
Time to first tenantScoped with you during the engagement. A standard OptiML Cloud tenant comes up in minutes; this one does not, because your account and network path have to be stood up first
Shared responsibilityYou own the availability, backup and patching of the database and object storage you host; RMT owns platform availability

6. Keys you hold, and can revoke

Custody is the part of a security review that cannot be argued around. Either you can make the data unreadable without asking us, or you cannot.

Recordings are encrypted with AES-256-GCM envelope encryption: a key per recording, wrapped by a tenant key held in KMS. Bring your own key and the tenant key is yours. Rotate it or revoke it and the data is unreadable without your action, which is a shorter answer to give a regulator than a paragraph about our internal access controls. TLS in transit, mTLS between services. PII is detected and redacted across transcripts, logs, exports and analytics before it reaches anyone's screen or anyone's report.

  • AES-256-GCM envelope encryption, a key per recording
  • Bring your own tenant key into KMS and it stays yours to rotate or revoke
  • TLS in transit, mTLS between services
  • PII redacted before it reaches a screen or a report, across transcripts, logs, exports and analytics

7. What 99.9% is made of

The SLA carries a 99.9% monthly uptime commitment with a published credit schedule, so a missed month has a consequence written down in advance. The list below is what that number is actually made of. Most of it is unglamorous plumbing, the queues and the autoscaling and the second region that can take the traffic. The drills matter more. A backup nobody has restored is a hope, and a failover nobody has triggered is a diagram. Incident history and subscribable updates sit on a public status page, which stays up whether the month went well or badly.

  • Microservices behind an event bus
  • Queues with dead-letter handling and idempotency
  • Horizontal autoscaling per service
  • Multi-region failover with replication
  • Automated backups, plus restores that someone has actually tested
  • Restore drills inside the release process, not once a year
  • Chaos experiments against production-like environments
  • A public status page with incident history

Documented RTO and RPO

RTO and RPO are written down. The current figures and the date of the last verified restore travel with the SLA and the security documentation. Alongside the status page, those are the only claims in this section a buyer can check without taking our word for anything, which is exactly why they are the ones worth putting in writing.

Request the SLA and restore evidence

8. Managed does not mean opaque

A Terraform provider covers agents, flows, knowledge sources, phone numbers, integrations, users and roles, so provisioning goes through review like any other change. Around it sit a REST API with OpenAPI, 11 client SDKs, a CLI, HMAC-signed webhooks with retries and a dead-letter queue, and an MCP server SDK. You configure and automate. We keep it running. Nothing in your configuration lives somewhere only RMT can reach.

  • Terraform provider for agents, flows, knowledge sources, phone numbers, integrations, users and roles
  • REST API with OpenAPI, plus 11 client SDKs
  • A CLI
  • HMAC-signed webhooks with retries and a dead-letter queue
  • MCP server SDK

What the Terraform provider provisions in Agent & Flow Builder

9. Operated by RMT

Everything in this table is a headcount decision you do not have to make. The platform you would otherwise be staffing an ops rota for is staffed by ours, and the telemetry still lands in your monitoring stack.

What RMT operates on your behalf:

Operated by RMT
ProvisioningA tenant in minutes; new regions without a project
Patching and upgradesContinuous, on our release train, with no maintenance weekend on your calendar
Monitoring and on-call24×7 platform monitoring; up to 24×7 contracted response on Enterprise
ScalingAutoscaling per service; you do not size anything
Backup and DRAutomated backups, verified restores, documented RTO and RPO, restore drills
Security operationsVulnerability management, dependency scanning, 50+ CI/CD quality and security gates
Telemetry to youOpenTelemetry traces, Prometheus metrics and structured PII-redacted logs exported into your Datadog, Honeycomb or Sentry

10. What we do not do

Can we run OptiML inside your own perimeter? No. OptiML is a managed cloud service, and that is a deliberate choice rather than a gap in the roadmap. It is why upgrades are continuous, why the release train is a single train for every tenant, and why 99.9% is a number we can be held to. If your obligation is about where conversation data rests, customer-managed data residency answers it. If your obligation is about where the software executes, then we are not the right vendor for you. Better you find that out here than in month six.

Use Cases

Where Deployment & Hosting delivers value

Tenant isolation enforced by the database

A client audit that stops at one programme

An outsourcer running eleven client programmes (illustrative)

Scenario

Each client programme is provisioned as its own tenant, with its own retention window, compliance mode, audit trail and DNC list. Then one client's auditor asks the three questions auditors always ask: what was stored, who reached it, when was it erased. The answer is scoped to that tenant, and row-level security on 231 tables is what makes that scope real.

Outcome

The audit is answered per programme instead of per platform. A defect in application code cannot pull another client's rows into the response, because the application is not what enforces the boundary.

Customer-managed data residency

The database never leaves the bank

A regional bank under a domestic data mandate (illustrative)

Scenario

The obligation is about where conversation data rests, not where the software executes. So the PostgreSQL database and the S3-compatible recording store are stood up in the bank's own cloud account, reached over a private, mutually authenticated connection. Everything else stays put. Application, AI runtime, media plane and consoles keep running in OptiML Cloud.

Outcome

The only durable copy of transcripts and recordings sits in an account the bank controls, under a tenant key in its own KMS. Revoke the key and the data is unreadable, without anyone having to raise a request with RMT first.

Configuration as code

A seasonal campaign ships as a pull request

The in-house support team at a subscription retailer (illustrative)

Scenario

Agents, flows, knowledge sources, phone numbers and role assignments are declared in the Terraform provider and reviewed like any other change. The festive-season campaign is a branch. Two new flows, one extra number, a wider set of knowledge sources, promoted from staging to production once someone has approved the diff.

Outcome

Configuration is versioned, diffable and revertable. Going live does not depend on somebody at RMT clicking things in the right order on the right day, and rolling back is the same merge request in reverse.

At a glance

Specification

The numbers and limits, without the sales copy

Specification for Deployment & Hosting
Specification Detail
Hosting OptiML Cloud, multi-tenant, in your chosen region
Enterprise hosting options Dedicated single-tenant instance inside OptiML Cloud, own retention windows and compliance posture, no shared infrastructure
Time to first tenant Minutes on OptiML Cloud. Customer-managed data residency is scoped during the engagement, because your account and network path have to be stood up first
Data residency Regional storage. On Enterprise, customer-managed data residency puts the database and recording storage in your own cloud account or datacentre
Isolation PostgreSQL row-level security on 231 tables, fail-closed tenant context derived from the authenticated token
Encryption AES-256-GCM envelope encryption, a key per recording, customer-held tenant key in KMS, TLS in transit and mTLS between services
Uptime 99.9% monthly commitment in the SLA with published credits, multi-region failover, autoscaling, verified restores
Configuration as code Terraform provider (agents, flows, knowledge sources, phone numbers, integrations, users, roles), REST API with OpenAPI, 11 SDKs, CLI, HMAC-signed webhooks, MCP server SDK
Observability exported to you OpenTelemetry traces, Prometheus metrics, PII-redacted structured logs into Datadog, Honeycomb or Sentry
Compliance 16+ regional packs, DPA, sub-processor register, immutable audit with integrity proofs
Not offered On-premises or customer-perimeter installation of the platform itself
FAQ

Questions,
answered

What teams ask us before they roll out Deployment & Hosting — how it works, what it needs from your side, and what happens when it gets something wrong

Still not sure?

Talk to a specialist and get a straight answer.

Ask our team

An OptiML Cloud tenant needs three things from you: a region, a retention window for each class of data, and the numbers and integrations you want connected. The tenant itself comes up in minutes. Customer-managed data residency asks for rather more, because you are hosting the data plane: a cloud account or datacentre, a network path for the private connection, and a KMS to hold the tenant key. That timeline gets scoped with you during the engagement, because we would only be guessing if we printed a number here. The long pole is almost never our side of the work but the internal approval to open a private network path between your cloud account and ours.

OptiML enforces tenant isolation inside PostgreSQL, using row-level security policies on 231 tables, so the database refuses to return another tenant's row whatever the application asks for. Tenant context comes from the authenticated token and fails closed. If nothing resolves, nothing comes back. Your security reviewer can test that directly. It is a control, not a convention.

Under customer-managed data residency, your database and recording storage hold the only durable copy of conversations, transcripts and recordings. Both sit in your account, not ours. Live audio, transcripts and prompts transit OptiML Cloud while they are being processed, along with requests to whichever model providers you have enabled, and no primary copy of any of that is kept. What persists with RMT is PII-redacted operational telemetry and a bounded message-queue window, both enumerated by data class in the DPA.

RMT patches and upgrades OptiML continuously, on one release train, so there is no maintenance weekend on your calendar and no pinned version you will have to be migrated off in two years. The trade-off is worth naming. You do not choose when an upgrade lands. That is part of what makes the 99.9% commitment something RMT can stand behind, and it is the same reason there is only ever one version of the platform to support.

The OptiML SLA publishes a credit schedule, so a missed month carries a consequence agreed in advance. Behind the number sit the ordinary mechanisms of a distributed system: microservices on an event bus, queues that handle their own dead letters, autoscaling per service and multi-region failover with replication. The part worth asking about is narrower. Backups get restored and verified on the release cadence. Chaos experiments run against production-like environments. Incident history stays on the public status page whether the month was good or bad.

Connected solutions

Where Deployment & Hosting is used

The Solutions pages that lean on this module, and what it looks like once it is configured for a particular floor, job title or job to be done.

9 solutions built on this module
Talk to a specialist

Bring your security reviewer

Send the questionnaire before the call. The controls, the sub-processor register and the DPA answer most of it in writing, which usually leaves two or three questions about your own jurisdiction. Those are the ones worth an hour of somebody senior. The reference architecture and the shared-responsibility matrix are available under NDA.

  • Questionnaire answered in writing, before the call
  • Sub-processor register and DPA shared up front
  • Reference architecture under NDA

Book your slot

Leave your email and our team will come back to you within one business day.

or reach us directly

Your details stay private. We never share them.

This website uses cookies.

Cookies are small text files that allow us to create the best browsing experience for you on our site. By continuing to use this website or clicking "Accept & Close", you are agreeing to our use of cookies. To understand how we use cookies or how to manage them, please see our cookies policy.

Ask OptiML

Powered by RMT Engineering