See where FenxARC™ sits in your AI stack

Several products, assistants and agents call ARC. ARC applies the active list and policy, then reaches only the provider models, private endpoints and tools that list permits.

An agent submits work to ARC with an active list. ARC applies access and spend policy, then orchestrates only the models and tools approved for that list. Every tool result returns to ARC.

1. 01The agent invokes ARC inside one active list.
2. 02ARC applies access, privacy and limits before execution.
3. 03ARC calls only the models and tools that list exposes.
4. 04Each tool returns its result to ARC before the route continues.

Inside the trust boundary, the application reaches ARC and its privacy policy. ARC calls trusted handling tools and a trusted model. Outside the boundary, one external model receives only the shielded context needed for its part.

1. 01ARC classifies the request under your privacy policy.
2. 02Sensitive material is handled inside the trusted environment.
3. 03ARC builds the minimum context required for the external step.
4. 04The external model contributes its result to ARC.
5. 05ARC recombines the answer inside the boundary.

A production application calls ARC. ARC applies route policy and request-time limits, then selects from an ordered set of approved provider or private model paths.

1. 01ARC builds the permitted routes for this request.
2. 02The active objective orders those routes.
3. 03When a provider fails, the request takes the next path you set.
4. 04The application keeps the same ARC interface.

One customer network boundary contains the application, self-hosted ARC, policy, locally hosted models, approved tools and connected storage. There is no external path.