@mupt-ai/dari-gateway 0.1.0 → 0.2.1

package/src/models/provider_defaults.ts

@@ -1,4 +1,4 @@
-import { gatewayFor } from "./gateways";
+import { gatewayFor } from "./gateways.js";
 
 // Built-in model provider presentation and endpoint defaults; gateways carry
 // their own. Mirrors the hosted Dari model catalog's provider defaults; keep

package/src/models/runtime_overrides.ts

@@ -1,6 +1,6 @@
 import { canonicalModelId, isRecord } from "@mupt-ai/dari-router/policy-engine";
 
-import { gatewayFor } from "./gateways";
+import { gatewayFor } from "./gateways.js";
 
 const MODEL_CONTEXT_OVERRIDES: Record<string, number> = {
   // Selector and title models are not router candidate configs, so their

package/src/observability/errors.ts

@@ -2,9 +2,9 @@ import {
   gatewayPlatform,
   type OperationalErrorContext,
   type RequestErrorContext,
-} from "../platform/platform";
+} from "../platform/platform.js";
 
-export type { OperationalErrorCode } from "../platform/platform";
+export type { OperationalErrorCode } from "../platform/platform.js";
 
 export function captureRequestError(
   error: unknown,

package/src/platform/platform.ts

@@ -3,10 +3,15 @@ import type {
   ReasoningEffort,
   RoutingStrategy,
 } from "@mupt-ai/dari-router/policy-engine";
-import type { HarnessConversationIdentity } from "../runtime/harness_identity";
-import type { InteractionRequest, SourceProtocol } from "../interaction/types";
-import type { SubscriptionUsageLimit } from "../providers/subscription_limits";
-import type { ClientHarness, ProviderKeySource, ResolvedRouter } from "../types";
+import type { HarnessConversationIdentity } from "../runtime/harness_identity/index.js";
+import type { InteractionRequest, SourceProtocol } from "../interaction/types.js";
+import type { SubscriptionUsageLimit } from "../providers/subscription_limits.js";
+import type {
+  ClientHarness,
+  ProviderCredential,
+  ProviderKeySource,
+  ResolvedRouter,
+} from "../types.js";
 
 /**
  * Everything the gateway runtime needs from the deployment around it: router
@@ -24,6 +29,15 @@ export type GatewayPlatform = {
    * `prefixHashes`. Throws HttpError(401) for an unknown key.
    */
   resolveRouter(request: ResolveRouterRequest): Promise<ResolvedRouter>;
+  /**
+   * Optional. Supply the credential for one enabled model of this request,
+   * consulted after `resolveRouter` and before its provider keys (for the
+   * self-hosted gateway, its environment keys), so an embedder can serve
+   * per-tenant keys from its own store. Undefined keeps the resolved key.
+   */
+  resolveProviderCredential?(
+    request: ProviderCredentialRequest,
+  ): Promise<ProviderCredential | undefined>;
   /** Persist one completed turn's prefix entry and/or refresh a read entry. */
   writePrefixState(write: PrefixStateWrite): Promise<void>;
   /** Set or clear a conversation's prefetched routing lease. */
@@ -53,6 +67,15 @@ export type ResolveRouterRequest = {
   clientHarness?: ClientHarness;
 };
 
+export type ProviderCredentialRequest = {
+  provider: Provider;
+  model: string;
+  routerId: string;
+  organizationId: string;
+  // The gateway key the request authenticated with.
+  apiKey: string;
+};
+
 export type PrefixStateWrite = {
   organization_id: string;
   router_id: string;
@@ -94,8 +117,14 @@ export type RequestRecord = {
   isNewConversation: boolean;
   zdr: boolean;
   harnessConversationIdentity?: HarnessConversationIdentity;
+  // Parsed from the request's X-Dari-Tags header; empty when absent.
+  tags: RequestTags;
 };
 
+// Caller-supplied attribution labels (run, trial, setting), opaque to the
+// gateway.
+export type RequestTags = Record<string, string>;
+
 export type RequestSummary = {
   request_id: string;
   organization_id: string;
@@ -130,7 +159,10 @@ export type RequestSummary = {
   cache_write_tokens: number | null;
   output_tokens: number | null;
   total_tokens: number | null;
+  // Provider-reported cost when the provider returns one (OpenRouter, Vercel
+  // AI Gateway), otherwise priced from the model's rates when known.
   total_cost_usd: number | null;
+  cost_source: "provider" | "pricing" | null;
   pricing_known: boolean | null;
 };
 

package/src/protocol/input/anthropic_messages.ts

@@ -1,11 +1,11 @@
 import { isReasoningEffort, isRecord, type ReasoningEffort } from "@mupt-ai/dari-router/policy-engine";
 
-import { HttpError } from "../../http/errors";
-import { parseDataImageUrl } from "../../interaction/image";
-import { decodeUntrustedProviderContinuationState } from "../../interaction/provider_continuation_state";
-import { decodeUntrustedPortableReasoningState } from "../../interaction/portable_reasoning_state";
-import { toolsByWireName } from "../../interaction/tool_identity";
-import { explicitModelId, isAnthropicRouterModel, unsupportedModelMessage } from "./router_model";
+import { HttpError } from "../../http/errors.js";
+import { parseDataImageUrl } from "../../interaction/image.js";
+import { decodeUntrustedProviderContinuationState } from "../../interaction/provider_continuation_state.js";
+import { decodeUntrustedPortableReasoningState } from "../../interaction/portable_reasoning_state.js";
+import { toolsByWireName } from "../../interaction/tool_identity.js";
+import { explicitModelId, isAnthropicRouterModel, unsupportedModelMessage } from "./router_model.js";
 import type {
   ClientTool,
   InteractionContent,
@@ -17,7 +17,7 @@ import type {
   InteractionToolReference,
   ReasoningControls,
   ServerToolKind,
-} from "../../interaction/types";
+} from "../../interaction/types.js";
 
 const SUPPORTED_SERVER_TOOL_TYPES: readonly ServerToolKind[] = [
   "bash_20250124",

package/src/protocol/input/chat_completions.ts

@@ -5,7 +5,7 @@ import {
   type ResponseFormat,
 } from "@mupt-ai/dari-router/policy-engine";
 
-import { HttpError } from "../../http/errors";
+import { HttpError } from "../../http/errors.js";
 import type {
   InteractionContent,
   InteractionItem,
@@ -14,10 +14,10 @@ import type {
   InteractionResponseFormat,
   InteractionTool,
   InteractionToolChoice,
-} from "../../interaction/types";
-import { toolsByWireName } from "../../interaction/tool_identity";
-import { decodeUntrustedProviderContinuationState } from "../../interaction/provider_continuation_state";
-import { validateChatCompletionsRequest } from "./chat_completions_validation";
+} from "../../interaction/types.js";
+import { toolsByWireName } from "../../interaction/tool_identity.js";
+import { decodeUntrustedProviderContinuationState } from "../../interaction/provider_continuation_state.js";
+import { validateChatCompletionsRequest } from "./chat_completions_validation.js";
 
 export function chatCompletionsToInteraction(request: ChatCompletionRequest): InteractionRequest {
   validateChatCompletionsRequest(request);

package/src/protocol/input/chat_completions_validation.ts

@@ -1,12 +1,12 @@
 import { isReasoningEffort, isRecord, type ChatCompletionRequest } from "@mupt-ai/dari-router/policy-engine";
 
-import { HttpError } from "../../http/errors";
-import { parseDataImageUrl } from "../../interaction/image";
-import { CANONICAL_ROUTER_MODEL, explicitModelId, unsupportedModelMessage } from "./router_model";
+import { HttpError } from "../../http/errors.js";
+import { parseDataImageUrl } from "../../interaction/image.js";
+import { CANONICAL_ROUTER_MODEL, explicitModelId, unsupportedModelMessage } from "./router_model.js";
 import {
   isProtocolToolCallId,
   TOOL_CALL_ID_MAX_LENGTH,
-} from "../../interaction/tool_identity";
+} from "../../interaction/tool_identity.js";
 
 export function validateChatCompletionsRequest(request: ChatCompletionRequest): void {
   if (request.model !== CANONICAL_ROUTER_MODEL && explicitModelId(request.model) === null) {

package/src/protocol/input/openai_responses.ts

@@ -2,15 +2,15 @@ import { createHash } from "node:crypto";
 
 import { isReasoningEffort, isRecord, type ReasoningEffort } from "@mupt-ai/dari-router/policy-engine";
 
-import { HttpError } from "../../http/errors";
-import { parseDataImageUrl } from "../../interaction/image";
-import { decodeUntrustedProviderContinuationState } from "../../interaction/provider_continuation_state";
-import { CANONICAL_ROUTER_MODEL, explicitModelId, unsupportedModelMessage } from "./router_model";
+import { HttpError } from "../../http/errors.js";
+import { parseDataImageUrl } from "../../interaction/image.js";
+import { decodeUntrustedProviderContinuationState } from "../../interaction/provider_continuation_state.js";
+import { CANONICAL_ROUTER_MODEL, explicitModelId, unsupportedModelMessage } from "./router_model.js";
 import {
   isProtocolToolCallId,
   TOOL_CALL_ID_MAX_LENGTH,
   toolsByWireName,
-} from "../../interaction/tool_identity";
+} from "../../interaction/tool_identity.js";
 import type {
   InteractionContent,
   InteractionItem,
@@ -20,7 +20,7 @@ import type {
   InteractionTool,
   InteractionToolChoice,
   ToolKind,
-} from "../../interaction/types";
+} from "../../interaction/types.js";
 
 const UNSUPPORTED_INPUT_ITEM_TYPES = new Set([
   "additional_tools",

package/src/protocol/output/anthropic_messages.ts

@@ -1,13 +1,13 @@
 import type { RoutingDecision } from "@mupt-ai/dari-router/policy-engine";
 
-import type { InteractionOutput, InteractionRequest } from "../../interaction/types";
+import type { InteractionOutput, InteractionRequest } from "../../interaction/types.js";
 import {
   anthropicContent,
   anthropicMessageId,
   anthropicRouting,
   anthropicStopReason,
   anthropicUsage,
-} from "./anthropic_messages_output";
+} from "./anthropic_messages_output.js";
 
 export function anthropicMessagesResponse(args: {
   id: string;
@@ -27,6 +27,6 @@ export function anthropicMessagesResponse(args: {
     stop_sequence: null,
     usage: anthropicUsage(args.output),
     ...(args.output.safeguardResults ? { safeguard_results: args.output.safeguardResults } : {}),
-    dari_routing: anthropicRouting(args.routing, args.conversationId),
+    dari_routing: anthropicRouting(args.routing, args.id, args.conversationId),
   };
 }

package/src/protocol/output/anthropic_messages_output.ts

@@ -1,7 +1,7 @@
-import { parseDataImageUrl } from "../../interaction/image";
-import { encodeProviderContinuationState } from "../../interaction/provider_continuation_state";
-import { encodePortableReasoningState } from "../../interaction/portable_reasoning_state";
-import type { InteractionOutput } from "../../interaction/types";
+import { parseDataImageUrl } from "../../interaction/image.js";
+import { encodeProviderContinuationState } from "../../interaction/provider_continuation_state.js";
+import { encodePortableReasoningState } from "../../interaction/portable_reasoning_state.js";
+import type { InteractionOutput } from "../../interaction/types.js";
 import type { RoutingDecision } from "@mupt-ai/dari-router/policy-engine";
 
 export function anthropicMessageId(requestId: string): string {
@@ -100,6 +100,7 @@ export function anthropicUsage(output: Pick<InteractionOutput, "usage">): Record
 
 export function anthropicRouting(
   decision: RoutingDecision,
+  requestId: string,
   conversationId?: string
 ): Record<string, unknown> {
   return {
@@ -107,6 +108,7 @@ export function anthropicRouting(
     selected_model: decision.selectedModel,
     reasoning_effort: decision.reasoningEffort,
     reason: decision.reason,
+    request_id: requestId,
     ...(conversationId ? { conversation_id: conversationId } : {}),
     ...(decision.leaseTurnsRemaining === undefined
       ? {}

package/src/protocol/output/anthropic_messages_stream.ts

@@ -1,20 +1,20 @@
-import { piMessageToInteractionOutput } from "../../providers/pi_output";
-import { providerToolCallIdentity } from "../../interaction/tool_identity";
-import type { InteractionOutput, InteractionRequest, ProviderIdentity } from "../../interaction/types";
+import { piMessageToInteractionOutput } from "../../providers/pi_output.js";
+import { providerToolCallIdentity } from "../../interaction/tool_identity.js";
+import type { InteractionOutput, InteractionRequest, ProviderIdentity } from "../../interaction/types.js";
 import {
   isProviderFailureMessage,
   providerFailureFromEvent,
   providerFailureMessage,
   type ProviderStreamOutcome,
-} from "../../runtime/provider_stream_outcome";
-import { sseStream } from "./sse_stream";
+} from "../../runtime/provider_stream_outcome.js";
+import { sseStream } from "./sse_stream.js";
 import {
   anthropicMessageId,
   anthropicReasoningBlock,
   anthropicRouting,
   anthropicStopReason,
   anthropicUsage,
-} from "./anthropic_messages_output";
+} from "./anthropic_messages_output.js";
 
 export function anthropicMessagesSseStream(args: {
   id: string;
@@ -58,7 +58,7 @@ export function anthropicMessagesSseStream(args: {
           stop_reason: null,
           stop_sequence: null,
           usage: { input_tokens: 0, output_tokens: 0 },
-          dari_routing: anthropicRouting(args.routing, args.conversationId),
+          dari_routing: anthropicRouting(args.routing, args.id, args.conversationId),
         },
       });
 

package/src/protocol/output/chat_completions.ts

@@ -1,17 +1,17 @@
 import { isRecord, type ChatMessage, type RoutingDecision } from "@mupt-ai/dari-router/policy-engine";
 
-import { piMessageToInteractionOutput } from "../../providers/pi_output";
-import { providerToolCallIdentity } from "../../interaction/tool_identity";
-import { encodeProviderContinuationState } from "../../interaction/provider_continuation_state";
-import { chatReasoningDetailId } from "./chat_reasoning_detail";
-import type { InteractionOutput, InteractionRequest, ProviderContinuationState, ProviderIdentity } from "../../interaction/types";
+import { piMessageToInteractionOutput } from "../../providers/pi_output.js";
+import { providerToolCallIdentity } from "../../interaction/tool_identity.js";
+import { encodeProviderContinuationState } from "../../interaction/provider_continuation_state.js";
+import { chatReasoningDetailId } from "./chat_reasoning_detail.js";
+import type { InteractionOutput, InteractionRequest, ProviderContinuationState, ProviderIdentity } from "../../interaction/types.js";
 import {
   isProviderFailureMessage,
   providerFailureFromEvent,
   providerFailureMessage,
   type ProviderStreamOutcome,
-} from "../../runtime/provider_stream_outcome";
-import { sseStream } from "./sse_stream";
+} from "../../runtime/provider_stream_outcome.js";
+import { sseStream } from "./sse_stream.js";
 
 export function chatCompletionResponse(args: {
   id: string;
@@ -34,7 +34,7 @@ export function chatCompletionResponse(args: {
       },
     ],
     usage: usagePayload(args.output, args.modelCost),
-    dari_routing: routingPayload(args.routing, args.conversationId),
+    dari_routing: routingPayload(args.routing, args.id, args.conversationId),
   };
 }
 
@@ -75,7 +75,7 @@ export function chatCompletionsSseStream(args: {
           usage: usagePayload(message, args.modelCost?.(message, true)) });
       };
 
-      send(chunk({ completion, created, model: args.model, delta: {}, finishReason: null, routing: args.routing, conversationId: args.conversationId }));
+      send(chunk({ completion, created, model: args.model, delta: {}, finishReason: null, routing: routingPayload(args.routing, args.id, args.conversationId) }));
       send(chunk({ completion, created, model: args.model, delta: { role: "assistant" }, finishReason: null }));
 
       const argumentDeltaSeen = new Set<string>();
@@ -208,12 +208,13 @@ export function chatCompletionsSseStream(args: {
   });
 }
 
-function routingPayload(decision: RoutingDecision, conversationId?: string): Record<string, unknown> {
+function routingPayload(decision: RoutingDecision, requestId: string, conversationId?: string): Record<string, unknown> {
   return {
     requested_model: "dari/routing",
     selected_model: decision.selectedModel,
     reasoning_effort: decision.reasoningEffort,
     reason: decision.reason,
+    request_id: requestId,
     ...(conversationId ? { conversation_id: conversationId } : {}),
   };
 }
@@ -224,9 +225,8 @@ function chunk(args: {
   model: string;
   delta: Record<string, unknown>;
   finishReason: string | null;
-  routing?: RoutingDecision;
+  routing?: Record<string, unknown>;
   usage?: Record<string, unknown>;
-  conversationId?: string;
 }): Record<string, unknown> {
   return {
     id: args.completion,
@@ -234,7 +234,7 @@ function chunk(args: {
     created: args.created,
     model: args.model,
     choices: [{ index: 0, delta: args.delta, finish_reason: args.finishReason }],
-    ...(args.routing ? { dari_routing: routingPayload(args.routing, args.conversationId) } : {}),
+    ...(args.routing ? { dari_routing: args.routing } : {}),
     ...(args.usage ? { usage: args.usage } : {}),
   };
 }

package/src/protocol/output/chat_reasoning_detail.ts

@@ -1,4 +1,4 @@
-import type { ProviderContinuationState } from "../../interaction/types";
+import type { ProviderContinuationState } from "../../interaction/types.js";
 
 export function chatReasoningDetailId(
   continuation: ProviderContinuationState,

package/src/protocol/output/openai_responses.ts

@@ -4,8 +4,8 @@ import {
   openAIResponsesRouting,
   openAIResponsesTerminal,
   openAIResponsesUsage,
-} from "./openai_responses_output";
-import type { InteractionOutput, InteractionRequest } from "../../interaction/types";
+} from "./openai_responses_output.js";
+import type { InteractionOutput, InteractionRequest } from "../../interaction/types.js";
 
 export function openAIResponsesResponse(args: {
   id: string;
@@ -24,7 +24,7 @@ export function openAIResponsesResponse(args: {
     output: openAIResponsesOutputItems(args.output),
     usage: openAIResponsesUsage(args.output),
     parallel_tool_calls: args.request.parallelToolCalls ?? true,
-    dari_routing: openAIResponsesRouting(args.routing, args.conversationId),
+    dari_routing: openAIResponsesRouting(args.routing, args.id, args.conversationId),
   };
 }
 import type { RoutingDecision } from "@mupt-ai/dari-router/policy-engine";

package/src/protocol/output/openai_responses_output.ts

@@ -1,5 +1,5 @@
-import { encodeProviderContinuationState } from "../../interaction/provider_continuation_state";
-import type { InteractionOutput } from "../../interaction/types";
+import { encodeProviderContinuationState } from "../../interaction/provider_continuation_state.js";
+import type { InteractionOutput } from "../../interaction/types.js";
 
 export function openAIResponsesResponseId(requestId: string): string {
   return `resp_${requestId}`;
@@ -44,6 +44,7 @@ export function openAIResponsesUsage(output: Pick<InteractionOutput, "usage">):
 
 export function openAIResponsesRouting(
   decision: RoutingDecision,
+  requestId: string,
   conversationId?: string
 ): Record<string, unknown> {
   return {
@@ -51,6 +52,7 @@ export function openAIResponsesRouting(
     selected_model: decision.selectedModel,
     reasoning_effort: decision.reasoningEffort,
     reason: decision.reason,
+    request_id: requestId,
     ...(conversationId ? { conversation_id: conversationId } : {}),
   };
 }

package/src/protocol/output/openai_responses_stream.ts

@@ -5,21 +5,21 @@ import {
   openAIResponsesRouting,
   openAIResponsesTerminal,
   openAIResponsesUsage,
-} from "./openai_responses_output";
-import { piMessageToInteractionOutput } from "../../providers/pi_output";
+} from "./openai_responses_output.js";
+import { piMessageToInteractionOutput } from "../../providers/pi_output.js";
 import type {
   InteractionOutput,
   InteractionOutputItem,
   InteractionRequest,
   ProviderIdentity,
-} from "../../interaction/types";
+} from "../../interaction/types.js";
 import {
   isProviderFailureMessage,
   providerFailureFromEvent,
   providerFailureMessage,
   type ProviderStreamOutcome,
-} from "../../runtime/provider_stream_outcome";
-import { sseStream } from "./sse_stream";
+} from "../../runtime/provider_stream_outcome.js";
+import { sseStream } from "./sse_stream.js";
 
 type StreamArgs = {
   id: string;
@@ -600,7 +600,7 @@ function responseEnvelope(
     output,
     parallel_tool_calls: args.request.parallelToolCalls ?? true,
     ...(interaction ? { usage: openAIResponsesUsage(interaction) } : {}),
-    dari_routing: openAIResponsesRouting(args.routing, args.conversationId),
+    dari_routing: openAIResponsesRouting(args.routing, args.id, args.conversationId),
   };
 }
 

package/src/protocol/output/sse_stream.ts

@@ -1,11 +1,11 @@
-import { ProviderStreamLifecycle } from "../../runtime/provider_stream_lifecycle";
+import { ProviderStreamLifecycle } from "../../runtime/provider_stream_lifecycle.js";
 import {
   providerFailure,
   providerFailureFromCaught,
   providerFailureFromTerminal,
   type ProviderStreamOutcome,
-} from "../../runtime/provider_stream_outcome";
-import { DeliveredStreamSnapshot } from "./delivered_stream_snapshot";
+} from "../../runtime/provider_stream_outcome.js";
+import { DeliveredStreamSnapshot } from "./delivered_stream_snapshot.js";
 
 export type SseStreamLifecycleArgs = {
   onFinal?: (outcome: ProviderStreamOutcome) => void;

package/src/providers/anthropic_safeguards.ts

@@ -1,7 +1,7 @@
 import { isRecord } from "@mupt-ai/dari-router/policy-engine";
 
-import type { AnthropicSafeguards, InteractionRequest } from "../interaction/types";
-import type { ProviderFetch } from "./response_capture";
+import type { AnthropicSafeguards, InteractionRequest } from "../interaction/types.js";
+import type { ProviderFetch } from "./response_capture.js";
 
 // Claude Code's auto mode asks the Anthropic API to run its dangerous-tool-use
 // classifier server-side, at no charge, by sending `safeguards` with a paired

package/src/providers/aws_assume_role.ts

@@ -1,12 +1,12 @@
 import { AssumeRoleCommand, STSClient } from "@aws-sdk/client-sts";
 
-import { HttpError } from "../http/errors";
+import { HttpError } from "../http/errors.js";
 import type {
   AwsAssumeRoleProviderAuth,
   AwsSigV4ProviderAuth,
   ProviderAuth,
   ResolvedProviderAuth,
-} from "../types";
+} from "../types.js";
 
 const ROLE_SESSION_NAME = "dari-routing";
 // AssumeRole always goes through this regional STS endpoint. The customer's

package/src/providers/cancel_estimate.ts

@@ -8,7 +8,7 @@ import type {
   InteractionOutput,
   InteractionUsage,
   ProviderIdentity,
-} from "../interaction/types";
+} from "../interaction/types.js";
 
 /** Prose from current models averages about 4.8 characters per token. */
 export const ESTIMATED_CHARS_PER_TOKEN = 4.8;

package/src/providers/execution.ts

@@ -1,27 +1,27 @@
 import type { RoutingDecision } from "@mupt-ai/dari-router/policy-engine";
-import { HttpError } from "../http/errors";
-import { piMessageToInteractionOutput } from "./pi_output";
-import type { InteractionRequest } from "../interaction/types";
-import { captureOperationalError } from "../observability/errors";
+import { HttpError } from "../http/errors.js";
+import { piMessageToInteractionOutput } from "./pi_output.js";
+import type { InteractionRequest } from "../interaction/types.js";
+import { captureOperationalError } from "../observability/errors.js";
 import {
   CustomerDisconnectedError,
   customerCanceledProviderMessage,
-} from "../runtime/customer_disconnect";
+} from "../runtime/customer_disconnect.js";
 import {
   isProviderFailureMessage,
   providerFailureFromCaught,
   providerFailureFromTerminal,
   providerFailureMessage,
-} from "../runtime/provider_stream_outcome";
+} from "../runtime/provider_stream_outcome.js";
 import {
   reportedExecution,
   type ReportableProviderOutcome,
   type RoutedRuntimeReporter,
-} from "../runtime/runtime_reporter";
-import type { RoutedProtocol } from "../runtime/routed_protocol";
-import { trackBackgroundTelemetry } from "../telemetry/background";
-import { gatewayPlatform, type SubscriptionHold } from "../platform/platform";
-import { reportRoutingUsage } from "../telemetry/usage";
+} from "../runtime/runtime_reporter.js";
+import type { RoutedProtocol } from "../runtime/routed_protocol.js";
+import { trackBackgroundTelemetry } from "../telemetry/background.js";
+import { gatewayPlatform, type SubscriptionHold } from "../platform/platform.js";
+import { reportRoutingUsage } from "../telemetry/usage.js";
 import {
   prepareProviderExecution,
   primeProviderStream,
@@ -32,20 +32,20 @@ import {
   type ProviderExecution,
   type StartedCompletion,
   type StartedStream,
-} from "./provider_call";
+} from "./provider_call.js";
 import {
   runProviderAttemptPlan,
   type ProviderAttemptFailure,
   type ProviderAttemptPlanResult,
   type ProviderAttemptResult,
-} from "./provider_attempt_runner";
+} from "./provider_attempt_runner.js";
 import {
   estimatedCancelUsage,
   NO_DELIVERED_OUTPUT,
   observeDeliveredOutput,
-} from "./cancel_estimate";
-import { providerResponseCaptureFromError, type ProviderResponseCapture } from "./response_capture";
-import { createStreamFinalizer } from "./stream_finalization";
+} from "./cancel_estimate.js";
+import { providerResponseCaptureFromError, type ProviderResponseCapture } from "./response_capture.js";
+import { createStreamFinalizer } from "./stream_finalization.js";
 import {
   detectSubscriptionUsageLimit,
   failureWithSubscriptionUsageLimit,
@@ -53,7 +53,7 @@ import {
   subscriptionUsageLimitFromFailure,
   subscriptionSwitchedMessage,
   subscriptionUsageLimitMessage,
-} from "./subscription_limits";
+} from "./subscription_limits.js";
 
 export type ProviderChoice = {
   decision: RoutingDecision;