AGKit vs AIProxy
Choose AGKit when you want an agent runtime for actions inside your app, with model configuration, credentials, and spending controls managed in Cloud.
AIProxy provides split-key protection, DeviceCheck, endpoint restrictions, typed Swift integrations, and per-user or per-IP rate limits.
 AGKit | AIProxy | |
|---|---|---|
| Application stack | Swift, Android, React Native, and Web; one Cloud backbone | Typed Swift integrations and supported provider endpoints |
| Application actions | Agent runtime, tool contracts, permission hooks, and cancellation | Application decides how to execute model-selected actions |
| Configuration | Cloud-managed model tiers, credentials, and agent profiles | Provider-specific integrations and service configuration |
| Production controls | End-user identity, project/user budgets, and configured attestation | DeviceCheck, endpoint restrictions, user/IP limits |
| Application responsibility | Tool implementation, state validation, approvals, and actual undo | App owns its action runtime and recovery |
Choose your platform and access path for SDK availability and setup. Guard and confirmation support depends on the SDK execution path you integrate.
Where AIProxy fits
Your team wants provider-specific Swift integrations with managed key protection and abuse controls. You keep the application workflow and decide how model responses affect your product.
Where AGKit fits
You want application tools, the agent lifecycle, and Cloud operations together. Configure profiles, model tiers, credentials, and spending controls centrally, then connect the runtime to your app.
What your team implements
Your app supplies its tool handlers, state validation, and approval experience. Use the runtime hooks for the selected execution path and implement undo for changes your application can reverse.