@code-yeongyu/senpi-ai 2026.11.3-2 → 2026.11.3-3

package/dist/types.d.ts

@@ -213,6 +213,12 @@ export interface StreamOptions extends ProviderRequestOptions<Model<Api>> {
      * session-aware features. Ignored by providers that don't support it.
      */
     sessionId?: string;
+    /**
+     * Opaque, non-identifying routing key for the session. When set, Anthropic Messages sends it in
+     * the session-affinity header instead of `sessionId`, so a gateway keeps the session on one
+     * upstream account without learning the session id. Other providers ignore it.
+     */
+    sessionAffinityKey?: string;
     /**
      * Optional custom fields to merge into the outgoing provider request body.
      * Applied before `onPayload` so `onPayload` can observe/override them.

package/dist/utils/provider-diagnostic-vocabulary.d.ts

@@ -1,12 +1,16 @@
 import type { ProviderDiagnostic, ProviderDiagnosticCategory, ProviderDiagnosticEvidence } from "../provider-diagnostic.ts";
 /** Serialized UTF-8 budget for one emitted diagnostic. */
 export declare const PROVIDER_DIAGNOSTIC_MAX_BYTES = 512;
+/** A credential pool whose every account is limited until `resetsAt`, the soonest one is due back. */
+export declare function credentialPoolBlockedDiagnostic(resetsAt: number): ProviderDiagnostic | undefined;
 /**
  * Closed allowlist of provider error tokens and the family each one proves.
  * Exact tokens only: Anthropic `error.type` values, OpenAI-compatible
  * `error.code` values, and the precise OpenAI-compatible `error.type` values
  * that stand in for a missing code (`exceeded_current_quota_error`, senpi#4950).
  * `billing_error` proves a billing failure, not quota exhaustion, so it stays `unknown`.
+ * `credential_pool_blocked` is senpi's own token for a credential pool whose every account waits on a
+ * rate or usage limit (senpi#3975).
  */
 export declare function categoryForProviderCode(code: string): ProviderDiagnosticCategory | undefined;
 export declare function isTlsChainVerificationCode(code: string): boolean;

package/dist/utils/provider-diagnostic-vocabulary.js

@@ -1,12 +1,24 @@
 import { siwcErrorForCode } from "./provider-diagnostic-carrier.js";
 /** Serialized UTF-8 budget for one emitted diagnostic. */
 export const PROVIDER_DIAGNOSTIC_MAX_BYTES = 512;
+const CREDENTIAL_POOL_BLOCKED_CODE = "credential_pool_blocked";
+/** A credential pool whose every account is limited until `resetsAt`, the soonest one is due back. */
+export function credentialPoolBlockedDiagnostic(resetsAt) {
+    return buildProviderDiagnostic({
+        category: "rate_limit",
+        code: CREDENTIAL_POOL_BLOCKED_CODE,
+        evidence: "structured_code",
+        resetsAt,
+    });
+}
 /**
  * Closed allowlist of provider error tokens and the family each one proves.
  * Exact tokens only: Anthropic `error.type` values, OpenAI-compatible
  * `error.code` values, and the precise OpenAI-compatible `error.type` values
  * that stand in for a missing code (`exceeded_current_quota_error`, senpi#4950).
  * `billing_error` proves a billing failure, not quota exhaustion, so it stays `unknown`.
+ * `credential_pool_blocked` is senpi's own token for a credential pool whose every account waits on a
+ * rate or usage limit (senpi#3975).
  */
 export function categoryForProviderCode(code) {
     const siwc = siwcErrorForCode(code);
@@ -21,6 +33,7 @@ export function categoryForProviderCode(code) {
         // `account_pool_blocked`: a subscription account pool that refused the request locally until its soonest reset (senpi#4672).
         case "rate_limit_error":
         case "rate_limit_exceeded":
+        case CREDENTIAL_POOL_BLOCKED_CODE:
         case "account_pool_blocked":
             return "rate_limit";
         case "insufficient_quota":
@@ -97,6 +110,7 @@ export function isProviderCodeStatusCompatible(code, status) {
             return status === 401;
         case "rate_limit_error":
         case "rate_limit_exceeded":
+        case CREDENTIAL_POOL_BLOCKED_CODE:
         case "insufficient_quota":
         case "exceeded_current_quota_error":
         case "usage_limit_reached":

package/dist/utils/provider-failure-description.d.ts

@@ -22,6 +22,8 @@ export declare function stripTurnRetrySuppressionPrefix(message: string): string
  */
 export declare function describeProviderFailureForUser(errorMessage: string | undefined, options?: ProviderStallDescriptionOptions & {
     usageLimit?: "account" | "model";
+    /** The account limit's reset as the caller knows it, used when the provider text names none. */
+    resetsAt?: string;
 }): string | undefined;
 /**
  * The turn-ending message for a provider module that can no longer be imported because the

package/dist/utils/provider-failure-description.js

@@ -84,7 +84,7 @@ export function describeProviderFailureForUser(errorMessage, options = {}) {
         const reset = /\bresets?\s+([^"}\n]+)/i
             .exec(message)?.[1]
             ?.trim()
-            .replace(/[.;]+$/, "");
+            .replace(/[.;]+$/, "") ?? options.resetsAt;
         return `The ${provider ? `${provider} ` : ""}account usage limit has been reached${reset ? ` (resets ${reset})` : ""}. Switch to another provider, or configure another provider in retry.fallbackChains.`;
     }
     const stall = describeProviderStallForUser(message, options);

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@code-yeongyu/senpi-ai",
-	"version": "2026.11.3-2",
+	"version": "2026.11.3-3",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -88,7 +88,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "0.127.0",
 		"@aws-sdk/client-bedrock-runtime": "3.1136.0",
-		"@earendil-works/pi-telemetry": "npm:@code-yeongyu/senpi-telemetry@2026.11.3-2",
+		"@earendil-works/pi-telemetry": "npm:@code-yeongyu/senpi-telemetry@2026.11.3-3",
 		"@google/genai": "2.23.0",
 		"@smithy/node-http-handler": "4.12.1",
 		"google-auth-library": "10.6.2",