We publish outcome figures only with the customer's written approval, so this slot stays empty until one is signed off. In the meantime the mechanism above is the honest version.
A platform your team configures, not one they run
REST API with OpenAPI, 11 client SDKs, a Terraform provider, an MCP server SDK and HMAC-signed webhooks. Everything configurable as code, and nothing for your team to keep alive.
- Sandbox environment
- 99.9% monthly uptime commitment
- Hosted in the region you choose
Illustrative interface. Latencies and provider states are sample data, not a performance claim. The screen shows the OptiML AI provider layer: a routing policy set to pick the cheapest provider clearing the quality floor, and a table of large-language-model and speech-to-text providers with their role, latency, which party the usage bills to and health state. One provider is degraded and traffic has been routed away from it. Traces, metrics and PII-redacted logs are exported to the customer's own observability tools.
Built to be configured, not run
The integration surface is wide on purpose and the operational surface is zero on purpose. Your team writes the configuration; RMT keeps the platform up.
What you care about
Each of these six concerns is answered with the interface that satisfies it. Nothing in this column is a roadmap item, and nothing in it requires a professional-services engagement to reach.
| Concern | What OptiML gives you |
|---|---|
| Integration surface | 100+ connectors, two-way CRM and ITSM sync, a universal REST adapter, webhooks with retries and a dead-letter queue |
| Configuration as code | Terraform provider covering agents, flows, knowledge sources, phone numbers, integrations, users and roles; REST API with OpenAPI, CLI, signed webhooks, sandbox environment |
| Observability | OpenTelemetry traces, Prometheus metrics and PII-redacted structured logs exported into your Datadog, Honeycomb or Sentry |
| Vendor risk | 11 LLM / 8 STT / 8 TTS providers, BYO-model and BYO-key, open standards-based WebRTC media |
| Data ownership | Regional hosting, customer-held encryption keys, and database and recording storage in your own cloud account on Enterprise |
| Change safety | Eval-gated prompt promotion, agent versioning with rollback, simulation before release |
terraform {
required_providers {
optiml = { source = "optiml/optiml", version = "~> 1.0" }
}
}
resource "optiml_knowledge_source" "product_docs" {
name = "Product documentation"
type = "url"
url = "https://docs.example.com"
crawl { depth = 2, render_js = true }
}
resource "optiml_agent" "support" {
name = "Support Agent"
type = "voice"
engine = "hybrid"
language = "en-IN"
system_prompt = file("${path.module}/prompts/support.txt")
knowledge_source_ids = [optiml_knowledge_source.product_docs.id]
guardrails {
pii = "redact"
grounding = "required"
injection = "block"
}
budget {
per_run_usd = 0.50
per_day_usd = 500
}
}
resource "optiml_phone_number" "main" {
number = "+911234567890"
agent_id = optiml_agent.support.id
}
Resources cover agents, agent versions, knowledge sources, phone numbers, SIP trunks, workflows, integrations, users, roles, API keys and webhooks.
What changes in your week
- No proprietary media stack to reverse-engineer when a call quality issue lands on your desk.
- Provisioning is Terraform, not a support ticket.
- Every model call is traced, so “why did the bill spike” has an answer in the platform.
- Nothing to patch, no upgrade weekend on your calendar, and no on-call rota for the contact centre.
The interfaces your team will actually use
Every one of these ships on every plan. Most teams start with the sandbox environment and the OpenAPI document.
API reference
REST API with OpenAPI, cursor pagination, idempotency keys and rate-limit headers. API referenceSDKs
Eleven client SDKs, typed, generated from the same OpenAPI document as the docs. SDKsTerraform provider
Agents, flows, knowledge sources, phone numbers, integrations, users and roles as code. TerraformWebhooks
HMAC-signed events with retries and a dead-letter queue you can drain yourself. WebhooksMCP server
Expose your own tools to the agent over the Model Context Protocol. MCP serverAI governance
Gates, caps, evals and audit — eval-gated prompt promotion and agent rollback. AI governanceWhat your team never has to carry
One release train, one operator, one pager — and it is not yours. The visibility still comes to you.
We operate it
Provisioning, patching, monitoring and scaling are RMT's, against a 99.9% monthly uptime commitment with published service credits.
You keep the visibility
OpenTelemetry traces, Prometheus metrics and PII-redacted structured logs export into your Datadog, Honeycomb or Sentry.
No model lock-in
11 LLM, 8 speech-to-text and 8 text-to-speech providers behind one adapter, with automatic failover and BYO-key.
Data where you need it
Regional hosting by default, and on Enterprise your database and recording storage in your own cloud account under keys you hold.
Objections you'll hear, answered
The third one is where a lot of evaluations end, and it should end there if that is genuinely your requirement. The answer below names the boundary and does not work around it.
“Another SaaS to integrate.”
It integrates the way your own services do: a REST API with OpenAPI, 11 client SDKs, signed webhooks with retries and a dead-letter queue, and a Terraform provider that keeps agents, flows, knowledge sources and permissions in version control. What it does not bring is a cluster to patch, an upgrade weekend or an on-call rota. RMT runs the platform against a 99.9% monthly uptime commitment and exports OpenTelemetry traces and Prometheus metrics into your Datadog or Sentry, so you keep the visibility without the pager.
“How do we avoid model lock-in?”
The provider layer is an adapter factory, so switching from one LLM to another is configuration. Automatic failover is already in place, a quality/cost router picks the cheapest provider clearing your quality floor, and BYO-key means the contract with the model vendor is yours.
“We'd rather run it in our own environment.”
You can't, and that is a deliberate choice, not a gap. One release train is why upgrades are continuous and why 99.9% is a target we can be held to. If your obligation is about where conversation data rests, customer-managed data residency answers it. That is an option on the Enterprise plan. Your PostgreSQL database and recording storage run in your own cloud account or your own datacentre, privately connected, under keys you hold.
If your obligation is about where the software executes, we are not the right vendor, and you should know that now rather than in month six.
What persists with us, precisely. Under customer-managed data residency your account holds the only durable copy of transcripts and recordings. PII-redacted operational telemetry and a bounded message-queue retention window persist with RMT, and the trust centre enumerates that residue by data class.
Where this goes next
The developer hub has the runnable version of everything on this page, and the trust centre has the control inventory your security review will ask for.
Customer quotes and outcome numbers need written approval from the named customer, so this slot stays visibly empty until one exists. An invented one would be worse than the gap.
The platform underneath
Everything on this page is the OptiML CX Platform configured for IT & Engineering, not a separate product. These are the modules it leans on.
See it handle one of your real conversations
Bring a call recording, a policy document or a WhatsApp thread from your own operation. We ground an agent in it and run it live on the call — not a canned demo.
- 30 minutes
- A working agent grounded in your content
- No slide deck unless you want one
Book a demo
30 minutes, on your own content. No slide deck unless you want one.
Your details stay private. We never share them.