@inferencesh/sdk 0.24.0 → 0.25.0

package/CHANGELOG.md

@@ -7,6 +7,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.25.0] - 2026-10-12
+
+### Added
+
+- Generated types follow the API: device authorization records the starting device and proximity (`DeviceAuthCodeInfo.device` with `client`, `client_version`, `proximity`, `scopes`), `DeviceAuthStatusSlowDown` and the poll `interval`, the `code_confirmation_required` and `invalid_scope` error codes, and `cached_tokens` / `cache_write_tokens` on LLM usage and text meta items.
+
+## [0.24.0] - 2026-10-11
+
 ### Added
 
 - Client tools can return files. A `ClientToolHandlerFn` may return `{ result: string; attachments?: FileRef[] }` (new `ClientToolResult` type) besides a string; the files are uploaded first with POST /files (`client.files.upload`, the chat's `uploadFile`) and sent as `attachments` with the result. Images reach the model as image input when it supports images. `submitToolResult` (agent api, `Agent`, `AgentsAPI`, the React chat actions) takes optional `attachments`, sent only when non-empty. `ToolResultRequest.attachments` is added by hand to the generated types until the next `make types` picks it up from the API. String returns work as before.

package/dist/types.d.ts

@@ -645,6 +645,12 @@ export interface DeviceAuthPollResponse {
      * credential comes with it.
      */
     admin_until?: string;
+    /**
+     * Interval is the polling interval in seconds the device is now held
+     * to, set with status slow_down (RFC 8628 3.5: each one raises it by
+     * five seconds).
+     */
+    interval?: number;
 }
 /**
  * AdminElevateRequest is the body of POST /device/auth/elevate: a platform
@@ -2801,7 +2807,7 @@ export interface DecideEntitlementRequestRequest {
  * and APIError.Code on the legacy envelope. Clients branch on these, so a
  * code is a named const here, where it generates into models and the SDKs.
  */
-export type ErrorCode = "invalid_request" | "validation_error" | "unauthorized" | "forbidden" | "not_found" | "conflict" | "name_conflict" | "already_exists" | "not_configured" | "method_not_allowed" | "rate_limited" | "internal_error" | "team_role_required" | "blocked_by_usage_policy" | "last_owner" | "account_deactivated" | "account_banned" | "person_required" | "otp_required" | "requires_sign_in" | "insufficient_scope" | "impersonation_reason_required" | "mcp_auth_expired" | "admin_session_required" | "admin_scopes_not_approved" | "cli_session_required" | "admin_required" | "same_admin_required" | "browser_session_required" | "invalid_ttl" | "admin_authenticator_required" | "limit_exceeded" | "feature_not_available" | "entitlement_unavailable" | "payment_required" | "payment_method_required" | "form_closed" | "already_submitted" | "already_claimed" | "already_entitled" | "request_open" | "invalid_code" | "already_processed" | "expired" | "app_retired" | "app_maintenance" | "invalid_transition" | "SESSION_NOT_FOUND" | "SESSION_EXPIRED" | "SESSION_ENDED" | "WORKER_LEASED" | "APP_MISMATCH" | "VERSION_MISMATCH" | "agents_disabled" | "remote_offline" | "remote_timeout" | "harness_not_drivable" | "harness_too_old";
+export type ErrorCode = "invalid_request" | "validation_error" | "unauthorized" | "forbidden" | "not_found" | "conflict" | "name_conflict" | "already_exists" | "not_configured" | "method_not_allowed" | "rate_limited" | "internal_error" | "team_role_required" | "blocked_by_usage_policy" | "last_owner" | "account_deactivated" | "account_banned" | "person_required" | "otp_required" | "requires_sign_in" | "insufficient_scope" | "invalid_scope" | "impersonation_reason_required" | "mcp_auth_expired" | "admin_session_required" | "admin_scopes_not_approved" | "cli_session_required" | "admin_required" | "same_admin_required" | "browser_session_required" | "invalid_ttl" | "admin_authenticator_required" | "limit_exceeded" | "feature_not_available" | "entitlement_unavailable" | "payment_required" | "payment_method_required" | "form_closed" | "already_submitted" | "already_claimed" | "already_entitled" | "request_open" | "invalid_code" | "already_processed" | "expired" | "code_confirmation_required" | "app_retired" | "app_maintenance" | "invalid_transition" | "SESSION_NOT_FOUND" | "SESSION_EXPIRED" | "SESSION_ENDED" | "WORKER_LEASED" | "APP_MISMATCH" | "VERSION_MISMATCH" | "agents_disabled" | "remote_offline" | "remote_timeout" | "harness_not_drivable" | "harness_too_old";
 export declare const ErrorCodeInvalidRequest: ErrorCode;
 export declare const ErrorCodeValidationError: ErrorCode;
 export declare const ErrorCodeUnauthorized: ErrorCode;
@@ -2860,6 +2866,12 @@ export declare const ErrorCodeRequiresSignIn: ErrorCode;
  * narrowed login granted less). The detail names the scope.
  */
 export declare const ErrorCodeInsufficientScope: ErrorCode;
+/**
+ * ErrorCodeInvalidScope (400): a grant request (an OAuth authorization,
+ * a device sign-in's approval) named only scopes this server does not
+ * know. The detail names them.
+ */
+export declare const ErrorCodeInvalidScope: ErrorCode;
 /**
  * ErrorCodeImpersonationReasonRequired (403): a platform admin named a
  * team they are not a member of without a live impersonation grant.
@@ -2938,6 +2950,11 @@ export declare const ErrorCodeRequestOpen: ErrorCode;
 export declare const ErrorCodeInvalidCode: ErrorCode;
 export declare const ErrorCodeAlreadyProcessed: ErrorCode;
 export declare const ErrorCodeExpired: ErrorCode;
+/**
+ * A device sign-in started away from the approver is approved only
+ * with the code typed from the device (400).
+ */
+export declare const ErrorCodeCodeConfirmationRequired: ErrorCode;
 /**
  * Run-time refusals. AppRetired (410): the app no longer runs.
  * AppMaintenance (503, Retry-After): the owner paused it; the message is
@@ -5022,6 +5039,12 @@ export interface ToolInvocationDTO extends BaseModelDTO, PermissionModelDTO {
 export interface MetaItem {
     type: MetaItemType;
     tokens?: number;
+    /**
+     * CachedTokens and CacheWriteTokens are the parts of Tokens (on an input
+     * text item) read from and written to the provider's prompt cache.
+     */
+    cached_tokens?: number;
+    cache_write_tokens?: number;
     width?: number;
     height?: number;
     resolution_mp?: number;
@@ -6288,11 +6311,16 @@ export declare const OpIsNull: FilterOperator;
 export declare const OpIsNotNull: FilterOperator;
 export declare const OpIsEmpty: FilterOperator;
 export declare const OpIsNotEmpty: FilterOperator;
-export type DeviceAuthStatus = "pending" | "approved" | "expired" | "denied" | "valid" | "invalid" | "loading";
+export type DeviceAuthStatus = "pending" | "approved" | "expired" | "denied" | "slow_down" | "valid" | "invalid" | "loading";
 export declare const DeviceAuthStatusPending: DeviceAuthStatus;
 export declare const DeviceAuthStatusApproved: DeviceAuthStatus;
 export declare const DeviceAuthStatusExpired: DeviceAuthStatus;
 export declare const DeviceAuthStatusDenied: DeviceAuthStatus;
+/**
+ * DeviceAuthStatusSlowDown: still pending, and the device polled sooner
+ * than its interval, which is now five seconds longer (RFC 8628 3.5).
+ */
+export declare const DeviceAuthStatusSlowDown: DeviceAuthStatus;
 export declare const DeviceAuthStatusValid: DeviceAuthStatus;
 export declare const DeviceAuthStatusInvalid: DeviceAuthStatus;
 export declare const DeviceAuthStatusLoading: DeviceAuthStatus;
@@ -7310,7 +7338,17 @@ export interface LLMUsage {
     stop_reason: string;
     time_to_first_token: number;
     tokens_per_second: number;
+    /**
+     * PromptTokens counts every input token, cached ones included
+     * (OpenAI, xAI and OpenRouter count them this way).
+     */
     prompt_tokens: number;
+    /**
+     * CachedTokens is how many of PromptTokens were read from the provider's
+     * prompt cache; CacheWriteTokens how many were written to it.
+     */
+    cached_tokens: number;
+    cache_write_tokens: number;
     completion_tokens: number;
     total_tokens: number;
     reasoning_tokens: number;

package/dist/types.js

@@ -434,6 +434,12 @@ export const ErrorCodeRequiresSignIn = "requires_sign_in";
  * narrowed login granted less). The detail names the scope.
  */
 export const ErrorCodeInsufficientScope = "insufficient_scope";
+/**
+ * ErrorCodeInvalidScope (400): a grant request (an OAuth authorization,
+ * a device sign-in's approval) named only scopes this server does not
+ * know. The detail names them.
+ */
+export const ErrorCodeInvalidScope = "invalid_scope";
 /**
  * ErrorCodeImpersonationReasonRequired (403): a platform admin named a
  * team they are not a member of without a live impersonation grant.
@@ -512,6 +518,11 @@ export const ErrorCodeRequestOpen = "request_open";
 export const ErrorCodeInvalidCode = "invalid_code";
 export const ErrorCodeAlreadyProcessed = "already_processed";
 export const ErrorCodeExpired = "expired";
+/**
+ * A device sign-in started away from the approver is approved only
+ * with the code typed from the device (400).
+ */
+export const ErrorCodeCodeConfirmationRequired = "code_confirmation_required";
 /**
  * Run-time refusals. AppRetired (410): the app no longer runs.
  * AppMaintenance (503, Retry-After): the owner paused it; the message is
@@ -1054,6 +1065,11 @@ export const DeviceAuthStatusPending = "pending";
 export const DeviceAuthStatusApproved = "approved";
 export const DeviceAuthStatusExpired = "expired";
 export const DeviceAuthStatusDenied = "denied";
+/**
+ * DeviceAuthStatusSlowDown: still pending, and the device polled sooner
+ * than its interval, which is now five seconds longer (RFC 8628 3.5).
+ */
+export const DeviceAuthStatusSlowDown = "slow_down";
 export const DeviceAuthStatusValid = "valid";
 export const DeviceAuthStatusInvalid = "invalid";
 export const DeviceAuthStatusLoading = "loading";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@inferencesh/sdk",
-  "version": "0.24.0",
+  "version": "0.25.0",
   "description": "Official JavaScript/TypeScript SDK for inference.sh - Run AI models with a simple API",
   "type": "module",
   "main": "dist/index.js",