Enterprise trust for model routing, Task Intelligence, and Agent OS.
Customers buying an AI gateway or intelligent coordination brain need more than model access. They need data boundaries, cost visibility, auditability, legal documents, and a clear separation between intelligence and execution.
Provider keys stay server-side, customer API keys are stored as hashes, and production access should go through customer backends.
Enterprise availability, support response, incident classification, measurement, exclusions, and service-credit boundaries.
Customer data categories, subprocessors, provider routing, deletion/export path, and enterprise data-processing terms.
Account data, request data, operational logs, secrets, customer controls, and support requests.
OneAI plans, routes, records, approves, and verifies. OneClaw, OpenClaw, bots, customer agents, or humans execute.
Order form, MSA, DPA, SLA, invoices, support expectations, and enterprise procurement package.
OneAI is not positioned as a blind AI relay. The commercial product is designed to expose request ownership, provider/model routing, usage, estimated cost, failures, audit events, and Agent OS execution records so customers can operate AI as infrastructure.
Send this package to procurement.
Use this trust package when an enterprise customer asks how OneAI handles security, data, execution boundaries, invoices, legal review, and Agent OS protocol integration.
Data flow overview
| Layer | Responsibility |
|---|---|
| Customer backend | Sends API requests with a OneAI API key. Browser-side secrets are not recommended. |
| OneAI API | Authenticates, applies policy, routes models/tasks, logs metadata, and records usage/cost. |
| Model provider | Processes selected model requests according to the provider and model configuration. |
| Agent OS ledger | Stores handoff contracts, approval state, proof callbacks, result callbacks, and review state. |
| Operator console | Shows customers, keys, usage, failures, audit events, billing state, and execution records. |
Control checklist
Hashed customer keys, revocation, prefixes, allowed IPs, scopes, rate limits, and budgets.
Provider/model selection, allowlists, fallback policy, cost guards, and model health checks.
Token usage, estimated model cost, margin visibility, maxCostUsd, and monthly plan limits.
Login, key creation, billing changes, failed requests, Agent OS proof/result, and operator review events.
Approval policy, proof policy, result ledger, and no direct external action inside OneAI.
SLA, DPA, privacy, invoices, terms, enterprise order forms, and manual sales path.