@mupt-ai/dari-gateway 0.1.0 → 0.2.1

package/src/telemetry/report.ts

@@ -14,34 +14,34 @@ import {
   type RoutingDecision,
   type RoutingStrategy,
 } from "@mupt-ai/dari-router/policy-engine";
-import type { ImageCapabilityPrunedModel } from "../context/routing_context";
-import { HttpError, errorCode } from "../http/errors";
-import { executionProviderForModel } from "../models/execution_provider";
+import type { ImageCapabilityPrunedModel } from "../context/routing_context.js";
+import { HttpError, errorCode } from "../http/errors.js";
+import { executionProviderForModel } from "../models/execution_provider.js";
 import {
   isBackgroundRequest,
   type InteractionOutput,
   type InteractionRequest,
   type InteractionUsage,
   type ProviderIdentity,
-} from "../interaction/types";
-import { captureOperationalError } from "../observability/errors";
-import { gatewayPlatform } from "../platform/platform";
+} from "../interaction/types.js";
+import { captureOperationalError } from "../observability/errors.js";
+import { gatewayPlatform, type RequestTags } from "../platform/platform.js";
 import type {
   ProviderAttemptRecord,
   ProviderAttemptReport,
   FrozenProviderTerminal,
   ReportableProviderOutcome,
   ReportedProviderExecution,
-} from "../runtime/runtime_reporter";
+} from "../runtime/runtime_reporter.js";
 import {
   selectorTraceFromError,
   type SelectorRun,
   type SelectorTrace,
-} from "../strategies/routing";
-import type { ResolvedRouter } from "../types";
-import type { HarnessConversationIdentity } from "../runtime/harness_identity";
-import { trackBackgroundTelemetry } from "./background";
-import { modelCost } from "./model_cost";
+} from "../strategies/routing.js";
+import type { ResolvedRouter } from "../types.js";
+import type { HarnessConversationIdentity } from "../runtime/harness_identity/index.js";
+import { trackBackgroundTelemetry } from "./background.js";
+import { modelCost } from "./model_cost.js";
 import {
   routingBillingProviderAttemptFacts,
   type FrozenRoutingBillingTerminal,
@@ -49,26 +49,26 @@ import {
   type RoutingBillingFinalizationPayload,
   type RoutingBillingProviderAttemptFacts,
   type RoutingBillingTerminalOutcome,
-} from "./billing_finalization";
-import { estimateBenchmarkReplay } from "./benchmark_replay";
-import { extractInteractionObservability } from "./requests/interaction_observability";
+} from "./billing_finalization.js";
+import { estimateBenchmarkReplay } from "./benchmark_replay.js";
+import { extractInteractionObservability } from "./requests/interaction_observability.js";
 import {
   costString,
   finiteNumber,
   isRecord,
   tokenCount,
-} from "./ingest_values";
+} from "./ingest_values.js";
 import {
   buildCompletedPrefixHit,
   leaseCountdownForPrefix,
   prefixEntryForIngest,
   reportPendingLease,
   reportPrefixState,
-} from "./prefix_state";
-import { subscriptionUsageLimitFromFailure } from "../providers/subscription_limits";
-import { scrubSecrets } from "./scrub";
-import { encodeRetainedInteractionRequest } from "./requests/retained_interaction_request";
-import { toolResultObservations, toolSkillEvents } from "./tools/observations";
+} from "./prefix_state.js";
+import { subscriptionUsageLimitFromFailure } from "../providers/subscription_limits.js";
+import { scrubSecrets } from "./scrub.js";
+import { encodeRetainedInteractionRequest } from "./requests/retained_interaction_request.js";
+import { toolResultObservations, toolSkillEvents } from "./tools/observations.js";
 
 // The decision's candidate_pruning audit: core cache-aware pruning, extended
 // with models excluded up front for missing hard capabilities (image input).
@@ -161,6 +161,7 @@ export class RequestReporter {
     private readonly args: {
       requestId: string;
       clientRequestId: string | null;
+      tags: RequestTags;
       credentials: ResolvedRouter;
       routerId: string;
       interaction: InteractionRequest;
@@ -428,6 +429,9 @@ export class RequestReporter {
           : null
       );
       const pricingKnown = selectedModel ? this.pricingKnown(selectedModel) : null;
+      const costSource = usage?.costSource === "provider"
+        ? "provider"
+        : pricingKnown === true ? "pricing" : null;
       const observability = extractInteractionObservability(a.interaction);
       const inputTokens = tokenCount(usage?.input) ?? 0;
       const retainedClientRequest = encodeRetainedInteractionRequest(
@@ -676,10 +680,10 @@ export class RequestReporter {
           cache_write_tokens: tokenCount(usage?.cacheWrite),
           output_tokens: tokenCount(usage?.output),
           total_tokens: tokenCount(usage?.totalTokens),
-          total_cost_usd:
-            pricingKnown === true && isRecord(usage?.cost)
-              ? finiteNumber(usage.cost.total)
-              : null,
+          total_cost_usd: costSource === null || !isRecord(usage?.cost)
+            ? null
+            : finiteNumber(usage.cost.total),
+          cost_source: costSource,
           pricing_known: pricingKnown,
         },
         distinctId:
@@ -690,6 +694,7 @@ export class RequestReporter {
         isNewConversation: a.isNewConversation,
         zdr: a.zdr,
         harnessConversationIdentity: a.harnessConversationIdentity,
+        tags: a.tags,
       });
     } catch (error) {
       console.error("routing report assembly failed", error);

package/src/telemetry/requests/interaction_observability.ts

@@ -5,7 +5,7 @@ import {
   type InteractionRequest,
   type InteractionToolChoice,
   type ProviderContinuationState,
-} from "../../interaction/types";
+} from "../../interaction/types.js";
 
 const SCHEMA_VERSION = 1 as const;
 

package/src/telemetry/requests/retained_interaction_request.ts

@@ -1,4 +1,4 @@
-import type { InteractionRequest } from "../../interaction/types";
+import type { InteractionRequest } from "../../interaction/types.js";
 
 export const RETAINED_INTERACTION_REQUEST_FORMAT = "interaction_request_v1" as const;
 

package/src/telemetry/skills/claude/catalog.ts

@@ -1,7 +1,7 @@
 import type {
   InteractionMessage,
   InteractionRequest,
-} from "../../../interaction/types";
+} from "../../../interaction/types.js";
 
 const MAX_CAPABILITY_ID_CHARS = 512;
 const MAX_SKILL_NAME_CHARS = 512;

package/src/telemetry/skills/claude/file_reads.ts

@@ -1,19 +1,19 @@
 import type {
   InteractionOutput,
   InteractionRequest,
-} from "../../../interaction/types";
-import { interactionToolInput } from "../../tools/input";
+} from "../../../interaction/types.js";
+import { interactionToolInput } from "../../tools/input.js";
 import {
   claudeSkillCapabilityId,
   claudeSystemSkillNames,
-} from "./catalog";
+} from "./catalog.js";
 import {
   normalizedAbsoluteSkillPath,
   skillPathSegments,
   type NormalizedSkillPath,
-} from "../paths";
-import type { SkillObservation } from "../observations";
-import { readPathsFromShellCommand } from "../shell_file_reads";
+} from "../paths.js";
+import type { SkillObservation } from "../observations.js";
+import { readPathsFromShellCommand } from "../shell_file_reads.js";
 
 const MAX_COMMAND_CHARS = 64 * 1024;
 const SKILL_FILE_NAME = "SKILL.md";

package/src/telemetry/skills/claude/observations.ts

@@ -5,16 +5,16 @@ import type {
   InteractionRequest,
   InteractionToolCall,
   InteractionToolResult,
-} from "../../../interaction/types";
+} from "../../../interaction/types.js";
 import type {
   SkillEvidenceSource,
   SkillObservation,
-} from "../observations";
+} from "../observations.js";
 import {
   claudeAvailableSkillNames,
   claudeSkillCapabilityId,
   normalizedClaudeSkillName,
-} from "./catalog";
+} from "./catalog.js";
 
 const MAX_SKILL_PATH_CHARS = 16 * 1024;
 const UNSAFE_DISPLAY_CHARACTERS = /[\u0000-\u001f\u007f-\u009f\u202a-\u202e\u2066-\u2069]/u;

package/src/telemetry/skills/codex/catalog.ts

@@ -1,12 +1,12 @@
-import type { InteractionRequest } from "../../../interaction/types";
-import { normalizedCodexSkillName } from "./identity";
+import type { InteractionRequest } from "../../../interaction/types.js";
+import { normalizedCodexSkillName } from "./identity.js";
 import {
   joinedSkillPath,
   normalizedAbsoluteSkillPath,
   skillPathBasename,
   type NormalizedSkillPath,
   type PathPlatform,
-} from "../paths";
+} from "../paths.js";
 
 const MAX_CATALOG_ENTRIES = 2_000;
 const SKILLS_OPEN = "<skills_instructions>";

package/src/telemetry/skills/codex/exec_file_reads.ts

@@ -1,7 +1,7 @@
-import type { InteractionOutput } from "../../../interaction/types";
-import { interactionToolInput } from "../../tools/input";
-import { readPathsFromShellCommand } from "../shell_file_reads";
-import type { CatalogSkill, CodexSkillCatalog } from "./catalog";
+import type { InteractionOutput } from "../../../interaction/types.js";
+import { interactionToolInput } from "../../tools/input.js";
+import { readPathsFromShellCommand } from "../shell_file_reads.js";
+import type { CatalogSkill, CodexSkillCatalog } from "./catalog.js";
 
 const MAX_COMMAND_CHARS = 64 * 1024;
 const MAX_SKILL_PATH_CHARS = 16 * 1024;

package/src/telemetry/skills/codex/explicit.ts

@@ -1,12 +1,12 @@
 import type {
   InteractionMessage,
   InteractionRequest,
-} from "../../../interaction/types";
-import type { SkillObservation } from "../observations";
+} from "../../../interaction/types.js";
+import type { SkillObservation } from "../observations.js";
 import {
   codexSkillCapabilityId,
   normalizedCodexSkillName,
-} from "./identity";
+} from "./identity.js";
 
 const MAX_SKILL_PATH_CHARS = 16 * 1024;
 const UNSAFE_DISPLAY_CHARACTERS = /[\u0000-\u001f\u007f-\u009f\u202a-\u202e\u2066-\u2069]/u;

package/src/telemetry/skills/codex/file_reads.ts

@@ -1,14 +1,14 @@
 import type {
   InteractionOutput,
   InteractionRequest,
-} from "../../../interaction/types";
+} from "../../../interaction/types.js";
 import {
   codexExecCommand,
   readSkillsFromCodexCommand,
-} from "./exec_file_reads";
-import { codexFileSkillCatalog } from "./catalog";
-import { codexSkillCapabilityId } from "./identity";
-import type { SkillObservation } from "../observations";
+} from "./exec_file_reads.js";
+import { codexFileSkillCatalog } from "./catalog.js";
+import { codexSkillCapabilityId } from "./identity.js";
+import type { SkillObservation } from "../observations.js";
 
 /** Match model-emitted Codex file reads to the exact current skill catalog. */
 export function codexAutomaticSkillObservations(

package/src/telemetry/skills/observations.ts

@@ -1,13 +1,13 @@
-import { isBackgroundRequest, type InteractionOutput, type InteractionRequest } from "../../interaction/types";
-import type { HarnessConversationIdentity } from "../../runtime/harness_identity";
+import { isBackgroundRequest, type InteractionOutput, type InteractionRequest } from "../../interaction/types.js";
+import type { HarnessConversationIdentity } from "../../runtime/harness_identity/index.js";
 import {
   piStyleAutomaticSkillObservations,
   piStyleExplicitSkillObservations,
-} from "./pi";
-import { claudeFileReadObservations } from "./claude/file_reads";
-import { claudeSkillObservations } from "./claude/observations";
-import { codexExplicitSkillObservations } from "./codex/explicit";
-import { codexAutomaticSkillObservations } from "./codex/file_reads";
+} from "./pi.js";
+import { claudeFileReadObservations } from "./claude/file_reads.js";
+import { claudeSkillObservations } from "./claude/observations.js";
+import { codexExplicitSkillObservations } from "./codex/explicit.js";
+import { codexAutomaticSkillObservations } from "./codex/file_reads.js";
 
 export type SkillEvidenceSource =
   | "skill_content_injection"

package/src/telemetry/skills/pi.ts

@@ -2,9 +2,9 @@ import type {
   InteractionMessage,
   InteractionOutput,
   InteractionRequest,
-} from "../../interaction/types";
-import { interactionToolInput } from "../tools/input";
-import type { SkillObservation } from "./observations";
+} from "../../interaction/types.js";
+import { interactionToolInput } from "../tools/input.js";
+import type { SkillObservation } from "./observations.js";
 
 const CAPABILITY_PREFIX = "unattributed:name:";
 const MAX_CAPABILITY_ID_CHARS = 512;

package/src/telemetry/skills/shell_file_reads.ts

@@ -3,13 +3,13 @@ import {
   resolvedSkillPath,
   type NormalizedSkillPath,
   type PathPlatform,
-} from "./paths";
+} from "./paths.js";
 import {
   executableCommand,
   restrictedShellSegments,
   trustedExecutableName,
   type ShellOperator,
-} from "./shell_words";
+} from "./shell_words.js";
 
 const MAX_SHELL_NESTING = 16;
 const NO_VALUE_FLAGS = new Set<string>();

package/src/telemetry/skills/shell_words.ts

@@ -1,6 +1,6 @@
 import { posix, win32 } from "node:path";
 
-import type { PathPlatform } from "./paths";
+import type { PathPlatform } from "./paths.js";
 
 const MAX_COMMAND_TOKENS = 4_096;
 const SHELL_ASSIGNMENT = /^[A-Za-z_][A-Za-z0-9_]*=/u;

package/src/telemetry/tools/input.ts

@@ -1,6 +1,6 @@
 import { isRecord } from "@mupt-ai/dari-router/policy-engine";
 
-import type { InteractionToolCall } from "../../interaction/types";
+import type { InteractionToolCall } from "../../interaction/types.js";
 
 export function interactionToolInput(
   item: InteractionToolCall,

package/src/telemetry/tools/observations.ts

@@ -1,23 +1,23 @@
 import { isRecord, stableStringify } from "@mupt-ai/dari-router/policy-engine";
 
-import { interactionPrefixChain } from "../../context/interaction_fingerprint";
+import { interactionPrefixChain } from "../../context/interaction_fingerprint.js";
 import type {
   InteractionContent,
   InteractionItem,
   InteractionOutput,
   InteractionRequest,
   InteractionToolResult,
-} from "../../interaction/types";
-import { isProtocolToolCallId } from "../../interaction/tool_identity";
-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 type { HarnessConversationIdentity } from "../../runtime/harness_identity";
+} from "../../interaction/types.js";
+import { isProtocolToolCallId } from "../../interaction/tool_identity.js";
+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 type { HarnessConversationIdentity } from "../../runtime/harness_identity/index.js";
 import {
   inputSkillObservations,
   outputSkillObservations,
-} from "../skills/observations";
+} from "../skills/observations.js";
 
 const MAX_TOOL_SKILL_EVENTS = 1000;
 

package/src/telemetry/usage.ts

@@ -1,17 +1,18 @@
 import type { Provider } from "@mupt-ai/dari-router/policy-engine";
 import { randomUUID } from "node:crypto";
 
-import { HttpError } from "../http/errors";
-import { executionProviderForModel, keySourceForModel } from "../models/execution_provider";
-import { captureOperationalError } from "../observability/errors";
-import { gatewayPlatform } from "../platform/platform";
-import type { ProviderKeySource, ResolvedRouter } from "../types";
+import { HttpError } from "../http/errors.js";
+import { executionProviderForModel, keySourceForModel } from "../models/execution_provider.js";
+import { captureOperationalError } from "../observability/errors.js";
+import { gatewayPlatform, type RequestTags } from "../platform/platform.js";
+import type { ProviderKeySource, ResolvedRouter } from "../types.js";
 
 export type RoutingUsageContext = {
   organizationId: string;
   routerId: string;
   provider: Provider;
   model: string;
+  tags: RequestTags;
   startedAt: Date;
 };
 
@@ -75,6 +76,7 @@ export function usageContextFor(args: {
   routerId: string;
   selectedModel: string;
   keySource: ProviderKeySource;
+  tags: RequestTags;
 }): RoutingUsageContext | undefined {
   if (!args.billable || args.keySource !== "managed") return undefined;
   if (args.credentials.credit_exhausted) throw creditExhaustedError();
@@ -83,6 +85,7 @@ export function usageContextFor(args: {
     routerId: args.routerId,
     provider: executionProviderForModel(args.credentials, args.selectedModel),
     model: args.selectedModel,
+    tags: args.tags,
     startedAt: new Date(),
   };
 }
@@ -143,6 +146,7 @@ function usagePayload(context: RoutingUsageContext, message: Record<string, unkn
     total_cost_usd: usd(cost.total) ?? "0",
     started_at: context.startedAt.toISOString(),
     ended_at: new Date().toISOString(),
+    tags: context.tags,
   };
 }
 

package/src/types.ts

@@ -42,6 +42,14 @@ export type AwsAssumeRoleProviderAuth = {
 // Authentication a provider call can use directly.
 export type ResolvedProviderAuth = BearerProviderAuth | AwsSigV4ProviderAuth;
 
+// A bearer credential an embedding platform supplies for one request through
+// GatewayPlatform.resolveProviderCredential, optionally with its own endpoint.
+export type ProviderCredential = {
+  apiKey: string;
+  baseUrl?: string;
+  headers?: Record<string, string>;
+};
+
 export type ProviderAuth = ResolvedProviderAuth | AwsAssumeRoleProviderAuth;
 
 export type AwsProviderAuth = AwsSigV4ProviderAuth | AwsAssumeRoleProviderAuth;
@@ -93,12 +101,19 @@ export type ResolvedRouter = {
   provider_keys: Partial<Record<Provider, string>>;
   provider_auth?: Partial<Record<Provider, AwsProviderAuth>>;
   provider_key_sources?: Partial<Record<Provider, ProviderKeySource>>;
+  // Model -> the platform-supplied credential that replaces the provider key
+  // above for this request. Filled from resolveProviderCredential.
+  provider_credentials?: Record<string, ProviderCredential>;
   // Execution provider -> the user's subscription credential serving it.
   subscription_credential_ids?: Partial<Record<Provider, string>>;
   zdr?: boolean;
   allow_long_context?: boolean;
   speculative_routing?: boolean;
   primary_retries?: number;
+  // Serve only the requested `dari/model/<id>` at the requested thinking
+  // level: no selector, lease, fallback, or personal subscription, and an
+  // error for anything else, whatever the client sends.
+  strict?: boolean;
   model_fallback_enabled?: boolean;
   fallback_requires_different_provider?: boolean;
   evals?: RouterEval[];