Introduction
Optiak is one operating layer between your AI applications and any downstream service, including model providers, MCP servers, document vaults, guardrails and optimization engines. It applies your organization's configured rules to every request as it happens, so every team gets control, choice and efficiency without building them into each application.
Why use Optiak
AI adoption moves faster than companies can run it. Each team picks its own providers, keys, safety checks and logs, and the company ends up with as many AI stacks as it has teams. That fragmentation is the root problem:
- Nothing is comparable. Four teams mean four logs, four cost lines and four definitions of "approved".
- Nothing compounds. What one team learns, the rest of the company doesn't.
- Nothing scales. One change in the market, such as a new model or a price change, has to be handled again by every team.
Optiak removes the fragmentation: every application uses the same control layer, the same rules and the same records. Once AI runs through one place, the company can control it.
Control and compliance
Control is where fragmentation hurts most: nobody can say what AI is running, which data leaves the company, or what happened after the fact. With Optiak, the organization sets these controls once, and they apply to every application.
Visibility
Every request from every application is recorded, and dashboards show usage and cost across teams.
Security
The organization approves models and sets data residency, retention and processing policies restricting available downstream service providers. Guardrails can block, redact or record personal data according to your policy.
Compliance
Each request generates a trace of every step it went through, and audit logs record every administrative change.
Enforcement
Guardrails block unsafe requests, network access rules reject unknown callers, and blocking budgets stop spending at the limit. All in one place and before any provider is called.
Choice and efficiency
The same layer also solves the problems that come next:
- No lock-in. Use any supported provider or your own models, with your own keys. Switching is a change of model ID, not a migration.
- Optimized cost. Smart Router picks the right model for each request, and Context Engine trims what it sends.
- Less fragile. Your applications call one OpenAI-compatible API, and fallback models keep requests flowing when a provider fails.
How it works
Keep your stack and add one operating layer. Your applications send their requests to Optiak instead of to each provider. Optiak enforces each request against your policies, adds the data it needs, protects what must be protected, picks the model, gives it the tools it needs and records every decision.
Your AI apps
Applies your rules to every request as it happens
Governance
Policy compliance · Budgets
Data retrieval
Data Integrations
Security
Guardrails · Network access
Optimization
Smart Router · Context Engine
Tools
MCP servers
Any model
Five steps, and only the first three need you
- 1
Set up your organization
Once, in the console
Provider keys, approved models, governance rules, data and tools.
- 2
Configure each application
Minutes per application
Pick its models, data, tools and Guardrails.
- 3
Connect your app
One URL change
Create an API key and point your app at Optiak.
- 4
Analyze
On every request
Cost, latency, tokens and a trace, in one place.
- 5
Optimize
Continuously
Tune routing, context and models without changing your apps.
The Quickstart walks you through the first three steps.
The organization enables, the application narrows
The first two steps follow one rule used across Optiak: the organization decides what is allowed, in Company setup, and each application turns on only what it needs, in its Configuration tab. An application can never use more than the organization allows; for example, it can enable only models the organization has enabled.
Core concepts
These are the terms used across the docs. Other pages link here instead of redefining them. Follow the links to learn more about each concept, but if you are new to Optiak we suggest first checking the Where to go next section.
| Term | Definition |
|---|---|
| Organization | Your company's account in Optiak. It holds members, groups, projects, provider keys, the models the organization allows and its organization-wide policies; see Organization and Projects. |
| Project | A group of related applications. Access to applications is granted per project; see Organization and Projects. |
| Application | A product, service or environment that sends AI traffic through Optiak. It belongs to one project and has its own API keys, enabled models and configuration; see Organization and Projects. |
| Gateway | The part of Optiak your applications call. It receives LLM requests and tool calls, applies the application's settings and forwards them to providers and tools; see Gateway overview. |
| MCP Gateway | The Gateway endpoint for tool calls, using the Model Context Protocol (MCP); see Tools. |
| Model ID | The public identifier of a model in requests, in the form provider/model, for example openai/gpt-6-luna. The special ID smart asks Smart Router to choose the model; see Models and Routing. |
| Smart Router | The Gateway capability that chooses a model for each request among the application's eligible models; see Smart Router. |
| Context Engine | The Gateway capability that fits the conversation and retrieved context to what the model needs, for example by compressing repeated content or keeping only recent messages; see Context Engine. |
| Guardrails | Checks on each request for prompt injection, personal data and toxic content. Depending on the application's policy, a detection is recorded, redacted or blocked; see Guardrails. |
| Budget | A monthly spending limit for the organization, a project, an application or an API key. Track only records spend without blocking; Block requests rejects new requests when the limit is reached. |
| Tools | Capabilities exposed by connected MCP servers that applications can call through the MCP Gateway; see Tools. |
| Data Integration | A source of documents, such as uploaded files, a Google Drive folder or a Notion workspace, that the Gateway can search to add context to a request; see Data Integrations. |
| Provider key | Your organization's key for a model provider, such as OpenAI. Optiak uses it to call the provider on your applications' behalf; see Connect Model APIs. |
| Application API key | The key your service sends to Optiak to authenticate requests for one application. It is shown once, when you create it; see Authentication. |
| Request | One call to the Gateway, such as a chat completion. Each request is recorded with its status, model, latency, tokens and cost, and with a trace of the steps it went through; see Application Observability. |
| Group | A named set of organization members that can receive role assignments together; see Members and Access. |
| Role assignment | A role given to a member or group at organization or project level. Roles decide what each person can see and change; see Members and Access. |
Where to go next
- Quickstart: send your first request through the Gateway in a few minutes.
- Gateway: everything that acts on a request, from models and Smart Router to Tools and Guardrails.
- Compliance: organization-wide controls for data residency, network access and audit logs.
- Observability: the dashboards and request views, and how each metric is calculated.
- Administration: organizations, projects, members, roles, keys and budgets.
- Reference: endpoints, parameters and errors of the Optiak API.