Some environments cannot ship prompts to a cloud, ever.
A behavioral routing layer that depends on cloud connectivity is not deployable in air-gapped clinical environments, defense and intelligence settings, financial trading floors with strict egress controls, or compliance-isolated subsidiaries of larger enterprises. The deployment fails the security review before the model ever runs a prompt.
Bridge Stone is the same Bridge routing engine, packaged on hardware that operates entirely on the customer's local network. Optional egress paths exist for audit export on the customer's schedule. No prompt and no model response ever leaves the device's local network without an explicit configured action.
Stone is intentionally a small product. 2,800+ devices deployed, each one purchased into a specific environment with specific compliance constraints. Stone is not a general-purpose consumer device. Customers scope deployment on a call.
Three steps from prompt to decision.
Device provisioned with deployment context
Each Stone unit ships pre-provisioned with the customer's deployment configuration: behavioral policy, context modules, audit-policy template, and any required Bridge Context modules. Provisioning happens at Brevor before shipment.
Local routing fully on-device
Once installed, every prompt routes through the on-device behavioral engine. Routing decisions, context loading, and audit-trail capture all happen locally. No prompt or response leaves the device's local network during routing.
Audit exports on customer schedule
Audit data accumulates locally with full retention. Export happens on the customer's schedule — to a customer-controlled destination via the optional egress path, or via physical media if the deployment is fully air-gapped. The customer controls when and what leaves.
Where Bridge Stone lives in production.
Air-gapped clinical environments
Health systems and research facilities operating in environments where patient or research data cannot route through cloud infrastructure under any condition.
Used by a regional health system's clinical-research wing for AI workflows on protected research data.
Defense and intelligence
Defense contractors and intelligence-adjacent operators with strict egress controls and FedRAMP High or equivalent requirements.
Used by a defense contractor for AI workflows on classified-adjacent operational data.
Financial trading floors
Trading desks and financial-services environments with egress controls inherited from market-data agreements and exchange compliance.
Used by a credit union's institutional-trading desk for AI workflows on order-flow data.
Compliance-isolated subsidiaries
Subsidiaries of larger enterprises operating under separate compliance regimes where shared cloud infrastructure is prohibited.
Used by an energy company's regulated-utility subsidiary for AI workflows on grid-operations data.
What ships in every deployment.
One endpoint between your application and any model.
curl https://stone.local/v1/route \
-H "Authorization: Bearer $STONE_DEVICE_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Summarize today'\''s anomaly reports from grid-ops.",
"context_profile": "utility_grid_ops_v2",
"audit_policy": "ferc_strict"
}'