cartethyia 2.7.9 → 2.8.1

package/src/console/ip-bans/routes.ts

@@ -10,6 +10,7 @@ import { Elysia, t, type AnyElysia } from "elysia";
 import { errorResponse, requireGlobalAdmin } from "../shared/errors";
 import type { IpAbuseStore } from "../../security/abuse";
 import type { ConsoleAccessResolver } from "../auth/access";
+import { BAN_DURATION_CHOICES } from "../api-keys/ban-constants";
 
 /** One active IP ban, as the console returns it. */
 export interface IpBanResponse {
@@ -26,17 +27,8 @@ export interface IpBanRoutesConfig {
   readonly store: IpAbuseStore;
 }
 
-const BAN_DURATION_CHOICES: readonly { readonly minutes: number | null; readonly label: string }[] = [
-  { minutes: 60, label: "1 hour" },
-  { minutes: 6 * 60, label: "6 hours" },
-  { minutes: 12 * 60, label: "12 hours" },
-  { minutes: 24 * 60, label: "1 day" },
-  { minutes: 7 * 24 * 60, label: "7 days" },
-  { minutes: null, label: "Permanent" },
-] as const;
-
-/** Parses a duration choice into minutes; null = permanent. */
-export function parseBanDuration(value: unknown): number | null | undefined {
+/** Parses a duration choice into minutes; null = permanent; undefined = invalid. */
+function parseBanDuration(value: unknown): number | null | undefined {
   if (value === null || value === undefined) return value === null ? null : undefined;
   const minutes = Number(value);
   if (!Number.isFinite(minutes) || minutes <= 0) return undefined;
@@ -68,8 +60,10 @@ export function createIpBanRoutes(config: IpBanRoutesConfig): AnyElysia {
           const ip = body.ip.trim();
           if (ip.length === 0)
             return errorResponse(new Error("ip is required"), set, "Invalid IP ban request");
-          const minutes = body.minutes === undefined || body.minutes === null ? null : Math.round(body.minutes);
-          if (minutes !== null && (!Number.isFinite(minutes) || minutes <= 0))
+          // Absent is permanent (the default), exactly like an explicit null;
+          // parseBanDuration maps those to null and rejects anything else.
+          const minutes = body.minutes === undefined ? null : parseBanDuration(body.minutes);
+          if (minutes === undefined)
             return errorResponse(new Error("minutes must be a positive integer or null"), set, "Invalid IP ban request");
           await store.ban(ip, minutes === null ? 0 : minutes * 60_000);
           return {

package/src/console/observability/live.ts

@@ -29,8 +29,8 @@ import type { NetworkPoolSelector } from "../../network/pool/selector";
 import { eq } from "drizzle-orm";
 import type { CartethyiaDatabase } from "../../persistence/db-handle";
 import { networkPools } from "../../persistence/schema";
-import { subscribePoolHealth } from "../../network/pool-health-machine";
-import { poolByteSnapshot } from "../../network/pool/byte-accounting";
+import { subscribePoolHealth } from "../../network/pool/health";
+import { poolByteSnapshot } from "../../network/pool/accounting";
 import { consoleSseResponse, createConsoleSseStream } from "./sse";
 import { subscribeRequestCompleted } from "../../observability/request-notify";
 

package/src/console/observability/store.ts

@@ -196,59 +196,6 @@ const TOP_N = 50;
  */
 const USAGE_READ_THROUGH_TTL_SECONDS = 30;
 
-/**
- * Nearest-rank percentile over a sorted array — kept for the JS-side
- * `summarizeUsagePerformance` (still used by callers that already hold a
- * bounded event list). The usage endpoints read percentiles from SQL via
- * `summarizeUsagePerformanceSql`, which is what the dashboard polls.
- */
-function percentile(sorted: readonly number[], fraction: number): number {
-  if (sorted.length === 0) return 0;
-  const index = Math.min(sorted.length - 1, Math.max(0, Math.ceil(fraction * sorted.length) - 1));
-  return Math.round(sorted[index] ?? 0);
-}
-
-interface UsagePerformanceSummary {
-  avgResolveMs: number;
-  avgDurationMs: number;
-  latencyP50Ms: number;
-  latencyP95Ms: number;
-  latencyP99Ms: number;
-  /** Events each series was computed from; see `UsageSummaryTotals`. */
-  latencySampleCount: number;
-  resolveSampleCount: number;
-}
-
-/** Summarizes routing delay and request latency for an already-filtered period. */
-export function summarizeUsagePerformance(
-  events: readonly Pick<TelemetryEventRecord, "latencyMs" | "resolveMs">[],
-): UsagePerformanceSummary {
-  const latencies: number[] = [];
-  let latencySum = 0;
-  let resolveSum = 0;
-  let resolveCount = 0;
-  for (const event of events) {
-    if (event.latencyMs !== null) {
-      latencies.push(event.latencyMs);
-      latencySum += event.latencyMs;
-    }
-    if (event.resolveMs !== null) {
-      resolveSum += event.resolveMs;
-      resolveCount += 1;
-    }
-  }
-  latencies.sort((a, b) => a - b);
-  return {
-    avgResolveMs: resolveCount > 0 ? resolveSum / resolveCount : 0,
-    avgDurationMs: latencies.length > 0 ? latencySum / latencies.length : 0,
-    latencyP50Ms: percentile(latencies, 0.5),
-    latencyP95Ms: percentile(latencies, 0.95),
-    latencyP99Ms: percentile(latencies, 0.99),
-    latencySampleCount: latencies.length,
-    resolveSampleCount: resolveCount,
-  };
-}
-
 /**
  * The console dimension as the SQL reader's grouping key.
  *

package/src/console/providers/catalog/contracts.ts

@@ -83,9 +83,6 @@ export interface ProviderAccountBulkActionResult {
   readonly globalSkipped: number;
 }
 
-/** Cap on one discovery-commit request (D29). */
-export const DISCOVER_MODEL_COMMIT_MAX = 100;
-
 /**
  * One model the provider's own listing advertised, projected for the Add-Model
  * picker. Read-only: `GET .../models/discover` writes nothing. `models`
@@ -610,6 +607,9 @@ export interface ProviderAccountResponse {
   providerId: string;
   tenantId: string | null;
   label: string;
+  /** Structured identity learned from an OAuth login, when the provider exposed one. */
+  accountEmail?: string;
+  accountName?: string;
   credentialKind: CredentialKind;
   credentialMode?: CredentialMode;
   /**
@@ -1056,6 +1056,10 @@ export interface ProviderRoutingResponse {
    * default to skipping because their transport can't route through an HTTP
    * CONNECT proxy; it stays a real, per-tenant-overridable setting. */
   readonly skipProxy: boolean;
+  /** How device-code OAuth flows choose their egress: `random` (default)
+   * routes start+poll through a randomly selected active network pool so the
+   * upstream sees a rotating IP; `direct` always dials direct egress. */
+  readonly deviceFlowProxy: "random" | "direct";
   /** Built-in API-key upstream User-Agent; OAuth and custom providers retain their own identity. */
   readonly userAgent: string;
   /** Operator-set custom User-Agent override; `null` = no override (native identity is used). */
@@ -1089,6 +1093,8 @@ export interface UpdateProviderRoutingRequest {
   readonly enabled?: boolean;
   /** When true, this provider's requests always dial direct. */
   readonly skipProxy?: boolean;
+  /** How device-code OAuth flows choose their egress; absent = keep current. */
+  readonly deviceFlowProxy?: "random" | "direct";
   /** Built-in API-key upstream User-Agent. */
   readonly userAgent?: string;
   /** Operator-set custom User-Agent override; `null` clears it back to native. */

package/src/console/providers/catalog/model-operations.ts

@@ -12,7 +12,6 @@ import type { AccessDecision } from "../../../security/access-control";
 import { ConsoleDomainError, requireScope, requireTenantAnyScope } from "../../shared/errors";
 import { GatewayError } from "../../../protocol/gateway-error";
 import {
-  DISCOVER_MODEL_COMMIT_MAX,
   MODEL_READ_SCOPES,
   MODEL_WRITE_SCOPES,
 } from "./contracts";
@@ -390,11 +389,11 @@ export function createModelCatalogOperations(config: ProviderCatalogConfig) {
           "modelIds must be non-empty model ids",
         );
       }
-      if (modelIds.length === 0 || modelIds.length > DISCOVER_MODEL_COMMIT_MAX) {
+      if (modelIds.length === 0) {
         throw new ConsoleDomainError(
           "invalid_request",
           400,
-          `modelIds must contain 1-${DISCOVER_MODEL_COMMIT_MAX} entries`,
+          "modelIds must contain at least one entry",
         );
       }
       try {

package/src/console/providers/catalog/provider-operations.ts

@@ -892,7 +892,7 @@ export function createProviderCatalogOperations(config: ProviderCatalogConfig) {
                     // dynamic-import: B-02 lazy graph — buddy device-binding materializer, buddy-family only.
                     // The buddy-identity graph is lazy-loaded: the device-binding
                     // materializer only runs for buddy-family accounts (B-02).
-                    ? (await import("../../../providers/integrations/buddy/buddy-identity"))
+                    ? (await import("../../../providers/integrations/buddy/headers"))
                         .materializeBuddyDeviceBinding(account.id, authState)
                     : undefined;
                 const effectiveAuthState = materialized ?? authState;

package/src/console/providers/catalog/routes.ts

@@ -5,7 +5,6 @@ import { SERVICE_KINDS, WIRE_FAMILIES } from "../../../protocol/canonical-model"
 import { PROBE_MODES, PROBE_REASONING_EFFORTS } from "../../../providers/discovery/discovery-types";
 import {
   ACCOUNT_STATUSES,
-  DISCOVER_MODEL_COMMIT_MAX,
   CREDENTIAL_KINDS,
   CREDENTIAL_MODES,
 } from "./contracts";
@@ -44,6 +43,7 @@ const compatibilityProfileSchema = t.Object({
       }),
     ),
   ),
+  service_kinds: t.Optional(t.Array(literalUnion(SERVICE_KINDS))),
   streaming_usage_mode: t.Optional(t.Union([t.Literal("include_usage"), t.Literal("none")])),
   structured_output: t.Optional(
     t.Object({
@@ -132,7 +132,7 @@ const registerModelsBody = t.Object({
   metadata: t.Optional(modelMetadataOverrideBody),
 });
 const discoverCommitBody = t.Object({
-  modelIds: t.Array(t.String(), { minItems: 1, maxItems: DISCOVER_MODEL_COMMIT_MAX }),
+  modelIds: t.Array(t.String(), { minItems: 1 }),
 });
 const probeModelBody = t.Object({
   modelId: t.String(),
@@ -140,6 +140,7 @@ const probeModelBody = t.Object({
   route: t.Optional(t.String()),
   wireFamily: t.Optional(t.String()),
   accountId: t.Optional(t.String()),
+  accountIds: t.Optional(t.Array(t.String(), { maxItems: 100 })),
   prompt: t.Optional(t.String()),
   reasoningEffort: t.Optional(literalUnion(PROBE_REASONING_EFFORTS)),
   maxOutputTokens: t.Optional(t.Integer({ minimum: 1, maximum: 131072 })),