@mupt-ai/dari-gateway 0.2.0 → 0.3.0

package/dist/types.d.ts

@@ -0,0 +1,97 @@
+import type { CustomRouterConfig, Provider, ReasoningEffort, RouterEval, RouterEvalScore, RouterModelPrice, RouterPrefixHit, RoutingStrategy } from "@mupt-ai/dari-router/policy-engine";
+export type ProviderKeySource = "user" | "managed" | "subscription";
+export type BearerProviderAuth = {
+    type: "bearer";
+    apiKey: string;
+};
+export type AwsSigV4ProviderAuth = {
+    type: "aws_sigv4";
+    access_key_id: string;
+    secret_access_key: string;
+    session_token?: string | null;
+    region: string;
+};
+export type AwsAssumeRoleProviderAuth = {
+    type: "aws_assume_role";
+    role_arn: string;
+    external_id: string;
+    region: string;
+};
+export type ResolvedProviderAuth = BearerProviderAuth | AwsSigV4ProviderAuth;
+export type SubscriptionKind = "openai_codex" | "anthropic_claude_code" | "meta_muse";
+export type ProviderCredential = {
+    apiKey: string;
+    baseUrl?: string;
+    headers?: Record<string, string>;
+    subscription?: SubscriptionKind;
+};
+export type ProviderAuth = ResolvedProviderAuth | AwsAssumeRoleProviderAuth;
+export type AwsProviderAuth = AwsSigV4ProviderAuth | AwsAssumeRoleProviderAuth;
+export type PublicEvalCatalogScore = RouterEvalScore & {
+    source_row_name?: string | null;
+    source_date?: string | null;
+    last_synced_at?: string | null;
+    sample_count?: number | null;
+};
+export type PublicEvalCatalogCard = Omit<RouterEval, "scores"> & {
+    source_id: string;
+    source_name: string;
+    source_url: string;
+    harness: string;
+    updated_at: string | null;
+    scores: PublicEvalCatalogScore[];
+};
+export type ClientHarness = "claude_code" | "muse_code";
+export type ResolvedRouterModelConfig = {
+    provider: string;
+    native_model_id: string;
+    api: string;
+    base_url?: string | null;
+    context_window?: number | null;
+    standard_context_limit?: number | null;
+    allow_long_context?: boolean;
+    max_output_tokens?: number | null;
+    supported_thinking_levels?: ReasoningEffort[] | null;
+    default_thinking_level?: ReasoningEffort | null;
+    average_output_tokens?: Partial<Record<ReasoningEffort, number>> | null;
+    cost: RouterModelPrice;
+    supports_managed_key?: boolean;
+    fast_mode?: boolean;
+    fast_mode_cost?: RouterModelPrice | null;
+    zdr?: boolean;
+};
+export type ResolvedRouter = {
+    router_id: string;
+    organization_id: string;
+    api_key_id: string | null;
+    attributed_user_id: string | null;
+    enabled_models: string[];
+    provider_keys: Partial<Record<Provider, string>>;
+    provider_auth?: Partial<Record<Provider, AwsProviderAuth>>;
+    provider_key_sources?: Partial<Record<Provider, ProviderKeySource>>;
+    provider_credentials?: Record<string, ProviderCredential>;
+    subscription_credential_ids?: Partial<Record<Provider, string>>;
+    zdr?: boolean;
+    allow_long_context?: boolean;
+    speculative_routing?: boolean;
+    primary_retries?: number;
+    strict?: boolean;
+    strict_providers?: string[];
+    model_fallback_enabled?: boolean;
+    fallback_requires_different_provider?: boolean;
+    evals?: RouterEval[];
+    eval_score_imputation?: boolean;
+    image_capability_filtering?: boolean;
+    lease_context_enabled?: boolean;
+    selector_image_input_enabled?: boolean;
+    public_eval_catalog?: PublicEvalCatalogCard[] | null;
+    prefix_hits?: RouterPrefixHit[];
+    routing_strategy?: RoutingStrategy;
+    custom_config?: CustomRouterConfig | null;
+    model_thinking_levels?: Record<string, ReasoningEffort[]>;
+    fast_models?: string[];
+    model_prices?: Record<string, RouterModelPrice>;
+    model_configs?: Record<string, ResolvedRouterModelConfig>;
+    credit_exhausted?: boolean;
+    autorouting_enabled?: boolean;
+};

package/dist/types.js

@@ -0,0 +1 @@
+export {};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mupt-ai/dari-gateway",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "description": "Self-hostable LLM gateway for coding agents and APIs: one endpoint for Anthropic Messages, OpenAI Chat Completions, and OpenAI Responses, with subscription pooling, failover, and prompt-cache-aware routing.",
   "license": "Apache-2.0",
   "repository": {
@@ -9,14 +9,27 @@
   },
   "type": "module",
   "exports": {
-    ".": "./src/index.ts",
-    "./bun": "./src/bun.ts",
-    "./*": "./src/*.ts"
+    ".": {
+      "bun": "./src/index.ts",
+      "types": "./dist/index.d.ts",
+      "default": "./dist/index.js"
+    },
+    "./bun": {
+      "bun": "./src/bun.ts",
+      "types": "./dist/bun.d.ts",
+      "default": "./dist/bun.js"
+    },
+    "./*": {
+      "bun": "./src/*.ts",
+      "types": "./dist/*.d.ts",
+      "default": "./dist/*.js"
+    }
   },
   "bin": {
-    "dari-gateway": "./src/self_host/main.ts"
+    "dari-gateway": "./dist/self_host/main.js"
   },
   "files": [
+    "dist",
     "src",
     "!src/**/*.test.ts",
     "!src/**/*.fixtures.ts",
@@ -29,12 +42,14 @@
   ],
   "scripts": {
     "start": "bun run src/self_host/main.ts",
+    "build": "rm -rf dist && tsc -p tsconfig.build.json",
     "typecheck": "tsc --noEmit",
-    "test": "bun test"
+    "test": "bun test",
+    "prepack": "bun run build"
   },
   "dependencies": {
     "@aws-sdk/client-sts": "3.1063.0",
-    "@mupt-ai/dari-router": "0.2.4",
+    "@mupt-ai/dari-router": "0.2.5",
     "@mupt-ai/pi-ai": "1.4.0-mupt.1"
   },
   "devDependencies": {

package/src/bun.ts

@@ -1,4 +1,4 @@
 // Bun-only entry points: the Bun.serve HTTP server and the self-hosted
 // gateway, whose state lives in bun:sqlite.
-export { installShutdownHandlers, startServer } from "./http/serve";
-export { startSelfHostedGateway } from "./self_host/main";
+export { installShutdownHandlers, startServer } from "./http/serve.js";
+export { startSelfHostedGateway } from "./self_host/main.js";

package/src/context/eval_catalog.ts

@@ -1,4 +1,4 @@
-import type { PublicEvalCatalogCard } from "../types";
+import type { PublicEvalCatalogCard } from "../types.js";
 import type { RouterEval } from "@mupt-ai/dari-router/policy-engine";
 
 // Convert public cards into hidden ratio-calibration inputs. Cards without a

package/src/context/interaction_fingerprint.ts

@@ -2,20 +2,20 @@ import { createHash } from "node:crypto";
 
 import { stableStringify } from "@mupt-ai/dari-router/policy-engine";
 
-import { anthropicMessagesToInteraction } from "../protocol/input/anthropic_messages";
-import { internalChatCompletionsToInteraction } from "../protocol/input/chat_completions";
-import { anthropicContent } from "../protocol/output/anthropic_messages_output";
-import { assistantMessagePayload } from "../protocol/output/chat_completions";
-import { compatibleProviderContinuation } from "../interaction/provider_continuation_state";
+import { anthropicMessagesToInteraction } from "../protocol/input/anthropic_messages.js";
+import { internalChatCompletionsToInteraction } from "../protocol/input/chat_completions.js";
+import { anthropicContent } from "../protocol/output/anthropic_messages_output.js";
+import { assistantMessagePayload } from "../protocol/output/chat_completions.js";
+import { compatibleProviderContinuation } from "../interaction/provider_continuation_state.js";
 import type {
   InteractionItem,
   InteractionOutput,
   InteractionRequest,
   ProviderContinuationState,
   ProviderIdentity,
-} from "../interaction/types";
-import { providerToolCallId, toolWireName } from "../interaction/tool_identity";
-import { ROUTING_PREFIX_HASHES_MAX_COUNT } from "../platform/routing_contract";
+} from "../interaction/types.js";
+import { providerToolCallId, toolWireName } from "../interaction/tool_identity.js";
+import { ROUTING_PREFIX_HASHES_MAX_COUNT } from "../platform/routing_contract.js";
 import {
   appendFingerprintContentText,
   canonicalInteractionFingerprintStep,
@@ -25,7 +25,7 @@ import {
   normalizedAssistantFingerprintContent,
   type InteractionFingerprintStep,
   type ReasoningFingerprint,
-} from "./interaction_fingerprint_format";
+} from "./interaction_fingerprint_format.js";
 
 const MAX_CONTINUATION_TARGETS = 16;
 

package/src/context/interaction_fingerprint_format.ts

@@ -5,7 +5,7 @@ import { stableStringify } from "@mupt-ai/dari-router/policy-engine";
 import type {
   InteractionContent,
   InteractionRequest,
-} from "../interaction/types";
+} from "../interaction/types.js";
 
 // Bump when canonical fingerprint semantics change so persisted hashes from
 // incompatible contracts remain disjoint.

package/src/context/interaction_lease_depth.ts

@@ -2,7 +2,7 @@ import type {
   InteractionContent,
   InteractionMessage,
   InteractionRequest,
-} from "../interaction/types";
+} from "../interaction/types.js";
 
 type AssistantGroup = {
   kind: "assistant";

package/src/context/pricing.ts

@@ -1,6 +1,6 @@
 import type { ModelPricing, PricingLookup } from "@mupt-ai/dari-router/policy-engine";
-import type { ResolvedRouter } from "../types";
-
+import { registeredModelPrice } from "../providers/pi.js";
+import type { ResolvedRouter } from "../types.js";
 
 export function pricingLookupForRouter(credentials: ResolvedRouter): PricingLookup {
   return (model: string): ModelPricing | null => modelPricingForRouter(credentials, model);
@@ -10,7 +10,10 @@ export function modelPricingForRouter(
   credentials: ResolvedRouter,
   model: string,
 ): ModelPricing | null {
-  const price = credentials.model_prices?.[model];
+  // A model the platform neither prices nor configures is served from Pi's
+  // catalog (see buildModel), so its catalog price is the one it is billed at.
+  const price = credentials.model_prices?.[model]
+    ?? (credentials.model_configs?.[model] === undefined ? registeredModelPrice(model) : null);
   if (!price || typeof price.input !== "number" || typeof price.output !== "number") return null;
   // Null cached rates mean the provider publishes none: bill cached prompt
   // buckets at the input rate, never for free.

package/src/context/provider_prompt_accounting.ts

@@ -4,15 +4,15 @@ import {
   type CandidatePromptAccounting,
 } from "@mupt-ai/dari-router/policy-engine";
 
-import { toolsByWireName } from "../interaction/tool_identity";
+import { toolsByWireName } from "../interaction/tool_identity.js";
 import type {
   InteractionContent,
   InteractionRequest,
   ProviderIdentity,
-} from "../interaction/types";
-import { buildContext } from "../providers/pi_context";
-import { isOpenAIResponsesApi } from "../providers/model_capabilities";
-import type { InteractionPrefixContext } from "./interaction_fingerprint";
+} from "../interaction/types.js";
+import { buildContext } from "../providers/pi_context.js";
+import { isOpenAIResponsesApi } from "../providers/model_capabilities.js";
+import type { InteractionPrefixContext } from "./interaction_fingerprint.js";
 
 type PromptMetrics = {
   blockDepth: number;

package/src/context/routing_context.ts

@@ -15,34 +15,34 @@ import {
 } from "@mupt-ai/dari-router/policy-engine";
 import { randomUUID } from "node:crypto";
 
-import { HttpError } from "../http/errors";
-import { effectiveContextWindow } from "../models/context_limits";
-import { keySourceForModel } from "../models/execution_provider";
-import { requiredCapabilitiesFor } from "../interaction/capabilities";
-import { providerIdentityFromModel } from "../interaction/provider_continuation_state";
-import { isBackgroundRequest, type InteractionRequest, type ProviderIdentity } from "../interaction/types";
-import { requireCandidateModels } from "../models/ids";
-import { modelMetadataLookup } from "../providers/pi";
+import { HttpError } from "../http/errors.js";
+import { effectiveContextWindow } from "../models/context_limits.js";
+import { keySourceForModel } from "../models/execution_provider.js";
+import { requiredCapabilitiesFor } from "../interaction/capabilities.js";
+import { providerIdentityFromModel } from "../interaction/provider_continuation_state.js";
+import { isBackgroundRequest, type InteractionRequest, type ProviderIdentity } from "../interaction/types.js";
+import { requireCandidateModels } from "../models/ids.js";
+import { modelMetadataLookup } from "../providers/pi.js";
 import {
   modelSupportsImageInput,
   modelSupportsHostedWebSearch,
   modelSupportsReasoningConfigurationUpdate,
   providerApiCanSerializeResponseFormat,
-} from "../providers/model_capabilities";
+} from "../providers/model_capabilities.js";
 import {
   selectorMessagesForInteraction,
   type SelectorImage,
-} from "../strategies/selector_projection";
-import type { ResolvedRouter } from "../types";
+} from "../strategies/selector_projection.js";
+import type { ResolvedRouter } from "../types.js";
 import {
   interactionPrefixContext,
   interactionPrefixLookupHashes,
   type InteractionPrefixContext,
-} from "./interaction_fingerprint";
-import { pricingLookupForRouter } from "./pricing";
-import { providerPromptAccounting } from "./provider_prompt_accounting";
-import { leaseHistoryFromPrefixHits } from "./lease_history";
-import { interactionLeaseFenceDepth } from "./interaction_lease_depth";
+} from "./interaction_fingerprint.js";
+import { pricingLookupForRouter } from "./pricing.js";
+import { providerPromptAccounting } from "./provider_prompt_accounting.js";
+import { leaseHistoryFromPrefixHits } from "./lease_history.js";
+import { interactionLeaseFenceDepth } from "./interaction_lease_depth.js";
 
 export type ImageCapabilityPrunedModel = {
   model: string;

package/src/http/auth.ts

@@ -1,6 +1,6 @@
 import { timingSafeEqual } from "node:crypto";
 
-import { HttpError } from "./errors";
+import { HttpError } from "./errors.js";
 
 export function bearerToken(headers: Headers): string {
   const value = headers.get("authorization") ?? "";

package/src/http/handler.ts

@@ -1,14 +1,15 @@
 import {
   handleResponses,
   handleResponsesWebSocketFallback,
-} from "../ingress/codex";
-import { handleModels } from "../ingress/models";
-import { handleChatCompletions } from "../ingress/chat_completions";
-import { handleAnthropicMessages } from "../ingress/anthropic_messages";
-import { bindGatewayPlatform, type GatewayPlatform } from "../platform/platform";
-import { DEFAULT_ROUTER_ALIAS } from "../platform/routing_contract";
-import type { SelectorBackend } from "../strategies/routing";
-import { errorResponse, HttpError } from "./errors";
+} from "../ingress/codex.js";
+import { handleModels } from "../ingress/models.js";
+import { handleMuseCodeModels } from "../ingress/muse_models.js";
+import { handleChatCompletions } from "../ingress/chat_completions.js";
+import { handleAnthropicMessages } from "../ingress/anthropic_messages.js";
+import { bindGatewayPlatform, type GatewayPlatform } from "../platform/platform.js";
+import { DEFAULT_ROUTER_ALIAS } from "../platform/routing_contract.js";
+import type { SelectorBackend } from "../strategies/routing.js";
+import { errorResponse, HttpError } from "./errors.js";
 
 export type GatewayHandlerOptions = {
   platform: GatewayPlatform;
@@ -31,6 +32,12 @@ export function createRequestHandler(options: GatewayHandlerOptions) {
     ) {
       return Response.json({ status: "ok" });
     }
+    // Muse Code reads its model list from its base URL's origin; `dari muse`
+    // forwards it to the router's own path.
+    const museModels = url.pathname.match(/^(?:\/([^/]+))?\/muse-code\/models$/);
+    if (request.method === "GET" && museModels !== null) {
+      return await handleMuseCodeModels(request, museModels[1] ?? DEFAULT_ROUTER_ALIAS);
+    }
     // Claude Code probes this first-party endpoint before its Messages call
     // when ANTHROPIC_BASE_URL points at a compatible gateway.
     if (request.method === "HEAD" && url.pathname === "/api/hello") {

package/src/http/serve.ts

@@ -1,6 +1,6 @@
-import { captureOperationalError } from "../observability/errors";
-import { drainBackgroundTelemetryFor } from "../telemetry/background";
-import { startHttp2Server } from "./http2_server";
+import { captureOperationalError } from "../observability/errors.js";
+import { drainBackgroundTelemetryFor } from "../telemetry/background.js";
+import { startHttp2Server } from "./http2_server.js";
 
 const GRACEFUL_SERVER_STOP_TIMEOUT_MS = 5_000;
 const FORCE_STOP_SETTLE_TIMEOUT_MS = 250;

package/src/index.ts

@@ -2,7 +2,7 @@
 // (`@mupt-ai/dari-gateway/<dir>/<module>`) for deployments that compose the
 // gateway with their own platform. This entry needs no Bun APIs; the Bun
 // server and the SQLite-backed self-hosted gateway live in `./bun`.
-export { createRequestHandler, type GatewayHandlerOptions } from "./http/handler";
+export { createRequestHandler, type GatewayHandlerOptions } from "./http/handler.js";
 export type {
   GatewayPlatform,
   OperationalErrorContext,
@@ -15,8 +15,16 @@ export type {
   RequestTags,
   ResolveRouterRequest,
   SubscriptionHold,
-} from "./platform/platform";
-export { createAutoRoutingSelectorBackend } from "./strategies/auto_routing_selector";
-export { createPrioritySelectorBackend } from "./strategies/priority_selector";
-export { createLlmSelectorBackend, type SelectorBackend } from "./strategies/routing";
-export type { ProviderCredential, ResolvedRouter } from "./types";
+} from "./platform/platform.js";
+export {
+  DEFAULT_CATALOG_PROVIDERS,
+  listModelCatalog,
+  type CatalogModel,
+  type CatalogModelPrice,
+  type CatalogPriceTier,
+  type ModelCatalogOptions,
+} from "./models/catalog.js";
+export { createAutoRoutingSelectorBackend } from "./strategies/auto_routing_selector.js";
+export { createPrioritySelectorBackend } from "./strategies/priority_selector.js";
+export { createLlmSelectorBackend, type SelectorBackend } from "./strategies/routing.js";
+export type { ProviderCredential, ResolvedRouter, SubscriptionKind } from "./types.js";

package/src/ingress/anthropic_messages.ts

@@ -1,18 +1,18 @@
 import { isRecord } from "@mupt-ai/dari-router/policy-engine";
-import { anthropicMessagesResponse } from "../protocol/output/anthropic_messages";
-import { anthropicMessagesSseStream } from "../protocol/output/anthropic_messages_stream";
-import { anthropicToken } from "../http/auth";
-import { errorResponseHeaders, normalizeError } from "../http/errors";
-import { captureRequestError } from "../observability/errors";
-import { anthropicMessagesToInteraction } from "../protocol/input/anthropic_messages";
-import { handleRoutedCompletion } from "../runtime/routed_completion";
-import type { SelectorBackend } from "../strategies/routing";
-import { claudeCodeRequestContext } from "./claude_code/request_context";
+import { anthropicMessagesResponse } from "../protocol/output/anthropic_messages.js";
+import { anthropicMessagesSseStream } from "../protocol/output/anthropic_messages_stream.js";
+import { anthropicToken } from "../http/auth.js";
+import { errorResponseHeaders, normalizeError } from "../http/errors.js";
+import { captureRequestError } from "../observability/errors.js";
+import { anthropicMessagesToInteraction } from "../protocol/input/anthropic_messages.js";
+import { handleRoutedCompletion } from "../runtime/routed_completion.js";
+import type { SelectorBackend } from "../strategies/routing.js";
+import { claudeCodeRequestContext } from "./claude_code/request_context.js";
 import {
   claudeCodeConversationIdentity,
-} from "../runtime/harness_identity";
-import { parseJsonRequest } from "./json";
-import type { AnthropicSafeguards } from "../interaction/types";
+} from "../runtime/harness_identity/index.js";
+import { parseJsonRequest } from "./json.js";
+import type { AnthropicSafeguards } from "../interaction/types.js";
 
 type AnthropicErrorType =
   | "authentication_error"

package/src/ingress/anthropic_models.ts

@@ -6,7 +6,7 @@
    is rejected, bare tier names classify). Fixed created_at keeps the response
    byte-stable. */
 
-import { ANTHROPIC_ROUTER_MODEL_TIERS } from "../protocol/input/router_model";
+import { ANTHROPIC_ROUTER_MODEL_TIERS } from "../protocol/input/router_model.js";
 
 const ROUTER_MODEL_CREATED_AT = "2026-01-01T00:00:00Z";
 

package/src/ingress/chat_completions.ts

@@ -1,16 +1,17 @@
 import type { ChatCompletionRequest } from "@mupt-ai/dari-router/policy-engine";
-import { bearerToken } from "../http/auth";
-import { errorResponse } from "../http/errors";
-import { captureRequestError } from "../observability/errors";
-import { chatCompletionResponse, chatCompletionsSseStream } from "../protocol/output/chat_completions";
-import { chatCompletionsToInteraction } from "../protocol/input/chat_completions";
-import { handleRoutedCompletion } from "../runtime/routed_completion";
-import type { SelectorBackend } from "../strategies/routing";
+import { bearerToken } from "../http/auth.js";
+import { errorResponse } from "../http/errors.js";
+import { captureRequestError } from "../observability/errors.js";
+import { chatCompletionResponse, chatCompletionsSseStream } from "../protocol/output/chat_completions.js";
+import { chatCompletionsToInteraction } from "../protocol/input/chat_completions.js";
+import { handleRoutedCompletion } from "../runtime/routed_completion.js";
+import type { SelectorBackend } from "../strategies/routing.js";
 import {
   chatCompletionsConversationIdentity,
   openCodeRequestContext,
-} from "../runtime/harness_identity";
-import { parseJsonRequest } from "./json";
+} from "../runtime/harness_identity/index.js";
+import { parseJsonRequest } from "./json.js";
+import { museCodeClientId } from "./muse_code.js";
 
 export async function handleChatCompletions(
   request: Request,
@@ -22,15 +23,21 @@ export async function handleChatCompletions(
     const payload = await parseJsonRequest<ChatCompletionRequest>(request);
     const interaction = chatCompletionsToInteraction({ ...payload, source_protocol: "openai_chat" });
     const agentContext = openCodeRequestContext(request.headers);
+    const museClientId = museCodeClientId(request.headers);
     return await handleRoutedCompletion({
       request,
       routerId,
       apiKey,
-      payload: agentContext === undefined ? interaction : { ...interaction, agentContext },
+      payload: {
+        ...interaction,
+        ...(agentContext === undefined ? {} : { agentContext }),
+        ...(museClientId === undefined ? {} : { museCodeClientId: museClientId }),
+      },
       responseBody: chatCompletionResponse,
       streamBody: chatCompletionsSseStream,
       errorResponse,
       selectorBackend,
+      clientHarness: museClientId === undefined ? undefined : "muse_code",
       harnessConversationIdentity: chatCompletionsConversationIdentity(
         request.headers,
         payload.prompt_cache_key,

package/src/ingress/claude_code/request_context.ts

@@ -1,5 +1,5 @@
-import { HttpError } from "../../http/errors";
-import type { AgentRequestContext, AgentRequestClass } from "../../interaction/types";
+import { HttpError } from "../../http/errors.js";
+import type { AgentRequestContext, AgentRequestClass } from "../../interaction/types.js";
 
 const REQUEST_CLASSES = new Set<AgentRequestClass>([
   "main", "subagent", "workflow", "auxiliary", "compaction", "side_reply",

package/src/ingress/codex.ts

@@ -1,17 +1,18 @@
-import { bearerToken } from "../http/auth";
-import { errorResponse } from "../http/errors";
-import { captureRequestError } from "../observability/errors";
-import { openAIResponsesResponse } from "../protocol/output/openai_responses";
-import { openAIResponsesSseStream } from "../protocol/output/openai_responses_stream";
-import { openAIResponsesToInteraction } from "../protocol/input/openai_responses";
-import { parseJsonRequest } from "./json";
-import { handleRoutedCompletion } from "../runtime/routed_completion";
-import type { SelectorBackend } from "../strategies/routing";
+import { bearerToken } from "../http/auth.js";
+import { errorResponse } from "../http/errors.js";
+import { captureRequestError } from "../observability/errors.js";
+import { openAIResponsesResponse } from "../protocol/output/openai_responses.js";
+import { openAIResponsesSseStream } from "../protocol/output/openai_responses_stream.js";
+import { openAIResponsesToInteraction } from "../protocol/input/openai_responses.js";
+import { parseJsonRequest } from "./json.js";
+import { museCodeClientId } from "./muse_code.js";
+import { handleRoutedCompletion } from "../runtime/routed_completion.js";
+import type { SelectorBackend } from "../strategies/routing.js";
 import {
   codexConversationIdentity,
   codexRequestPurpose,
   codexRequestContext,
-} from "../runtime/harness_identity";
+} from "../runtime/harness_identity/index.js";
 
 export async function handleResponses(
   request: Request,
@@ -24,10 +25,12 @@ export async function handleResponses(
     const interaction = openAIResponsesToInteraction(rawPayload);
     const requestPurpose = codexRequestPurpose(request.headers, rawPayload);
     const agentContext = codexRequestContext(request.headers, rawPayload);
+    const museClientId = museCodeClientId(request.headers);
     const payload = {
       ...interaction,
       ...(requestPurpose === undefined ? {} : { requestPurpose }),
       ...(agentContext === undefined ? {} : { agentContext }),
+      ...(museClientId === undefined ? {} : { museCodeClientId: museClientId }),
     };
     const harnessConversationIdentity = codexConversationIdentity(
       request.headers,
@@ -47,6 +50,7 @@ export async function handleResponses(
       streamBody: openAIResponsesSseStream,
       errorResponse,
       selectorBackend,
+      clientHarness: museClientId === undefined ? undefined : "muse_code",
       harnessConversationIdentity,
     });
   } catch (error) {

package/src/ingress/codex_models.ts

@@ -1,7 +1,7 @@
-import type { ResolvedRouter } from "../types";
-import { CANONICAL_ROUTER_MODEL, explicitModelSlug } from "../protocol/input/router_model";
-import { effectiveContextWindow } from "../models/context_limits";
-import { modelSupportsHostedWebSearch } from "../providers/model_capabilities";
+import type { ResolvedRouter } from "../types.js";
+import { CANONICAL_ROUTER_MODEL, explicitModelSlug } from "../protocol/input/router_model.js";
+import { effectiveContextWindow } from "../models/context_limits.js";
+import { modelSupportsHostedWebSearch } from "../providers/model_capabilities.js";
 
 const DEFAULT_CODEX_CONTEXT_WINDOW = 128_000;
 

package/src/ingress/json.ts

@@ -6,7 +6,7 @@ import {
   createZstdDecompress,
 } from "node:zlib";
 
-import { HttpError } from "../http/errors";
+import { HttpError } from "../http/errors.js";
 
 const MAX_ENCODED_JSON_REQUEST_BYTES = 64 * 1024 * 1024;
 const MAX_DECODED_JSON_REQUEST_BYTES = 64 * 1024 * 1024;