OPA starts from a simple idea: a company should be able to use artificial intelligence intensively without giving up control of its infrastructure, data and costs. The video translates that idea into an architecture. It does not present an abstract cloud service, but a coherent system installed close to the company’s actual use cases.

Infrastructure that joins the company network

The first visible element is the server rack. It represents OPA’s hardware foundation, sized and installed on site. Computation takes place in the environment approved by the company, where the data required for private use cases can remain. Hardware, compute capacity and models are selected according to the real requirement. The objective is not to oversize the installation, but to build capacity suited to the expected volumes, response times and applications.

OPA Core, the common engine

OPA Core sits above this infrastructure. This private gateway provides an interface compatible with modern language-model workflows and runs the models selected for the project. It centralises access to the inference engine and exposes a private API. Applications therefore do not have to depend directly on an external provider for every request: they gain an access point controlled by the company.

This shared layer is essential. It makes it possible to evolve hardware or models without rebuilding every business tool. It also supports a consistent access policy, usage observation and greater predictability of the resources consumed.

Tools built on the same foundation

The second layer brings together the interfaces used by teams. OPA Safe Companion provides private assistance inside Visual Studio Code. OPA Chat offers a local, multimodal conversational interface. OPA Commandor is designed to supervise agents running in controlled environments. These products use the same OPA Core engine, avoiding a separate infrastructure and integration for every new use case.

The video also shows the API connected directly to internal applications. OPA is therefore not limited to the tools presented. A company can connect business software, a document workflow, private search or a specific automation to the same foundation, within the boundaries defined for the project.

Sovereignty, controlled costs and selected energy

Sovereignty here means operational capacity: knowing where the model runs, where data flows and who controls access. Cost control comes from replacing part of variable token expenditure with sized and monitored infrastructure. Selected energy means that the compute power used is chosen according to the model and workload, rather than consumed without visibility on every remote call. These three dimensions define the product’s direction.

A presentation aligned with current maturity

OPA is currently a suite under construction, with some components validated as proofs of concept and others still in development. OPA Core, Safe Companion and OPA Chat have validated POC workflows. Commandor is in development. The video presents the target product architecture that connects these components; it does not describe all of them as production services already available. Production readiness is prepared with each client according to its environment, models, data and expected service level.

Explore an OPA deployment

Would you like to evaluate the hardware capacity, models or use cases suited to your organisation? We can start from a concrete requirement and define a proof-of-concept scope.

Discuss your infrastructure

Tom Cheniaux - rephrased using AI