cartethyia 2.6.3 → 2.6.5

package/package.json

@@ -1,96 +1,96 @@
-{
-  "name": "cartethyia",
-  "version": "2.6.3",
-  "description": "A powerful, self-hosted reverse proxy for OpenAI- and Anthropic-compatible AI clients, powered by native Bun and Elysia. It serves both protocols from a single server, routes every model across any configured provider, and translates requests and responses in real time — streaming and prompt caching included. Underneath it runs a full delivery engine: per-account health tracking, quota and cooldown enforcement, automatic failover across pools and providers, usage accounting, and telemetry. Point OpenCode, Cline, Claude Code, Codex, or any compatible client at Cartethyia and it keeps its native protocol while the gateway absorbs every provider-specific difference behind it. Runs on embedded SQLite or external PostgreSQL with optional Redis coordination, ships as a global CLI with a system-tray console, and needs only Bun 1.4+ alongside Node.js/npm.",
-  "license": "GPL-3.0-only",
-  "type": "module",
-  "packageManager": "bun@1.4.0",
-  "repository": {
-    "type": "git",
-    "url": "git+https://github.com/risunCodex/Cartethyia.git"
-  },
-  "homepage": "https://github.com/risunCodex/Cartethyia#readme",
-  "bugs": {
-    "url": "https://github.com/risunCodex/Cartethyia/issues"
-  },
-  "keywords": [
-    "ai",
-    "gateway",
-    "llm",
-    "bun",
-    "self-hosted"
-  ],
-  "bin": {
-    "cartethyia": "scripts/cli/bin/cartethyia.cjs",
-    "carte": "scripts/cli/bin/carte.cjs"
-  },
-  "files": [
-    "scripts/cli/",
-    "src/",
-    "migrations/",
-    "dist/dashboard/",
-    ".env.example",
-    "README.md"
-  ],
-  "workspaces": [
-    "dashboard"
-  ],
-  "engines": {
-    "bun": ">=1.4.0"
-  },
-  "scripts": {
-    "dev": "bun run dev:stack",
-    "dev:watch": "bun run scripts/dev/supervisor.ts",
-    "test": "bun run scripts/test/run.ts",
-    "build": "bun run scripts/build/build.ts",
-    "prepack": "bun run dashboard:build",
-    "setup": "bun run scripts/cli/setup.ts",
-    "start": "bun run scripts/cli/main.ts start",
-    "doctor": "bun run scripts/cli/doctor.ts",
-    "probe:egress": "bun run scripts/tools/probe-egress.ts",
-    "restart": "bun run scripts/build/restart.ts",
-    "build:aot": "bun run scripts/build/aot.ts",
-    "dev:stack": "concurrently -n BACKEND,DASHBOARD -c cyan,magenta --kill-others-on-fail \"bun run dev:backend\" \"bun run dashboard:dev\"",
-    "typecheck": "tsc --noEmit",
-    "test:watch": "bun run scripts/test/run.ts --watch",
-    "test-db:up": "docker compose -f docker-compose.test.yml up -d --wait postgres-test",
-    "dev:backend": "bun run --hot src/main.ts",
-    "dev:console-auth": "bun run scripts/test/console-session.ts",
-    "dev:visual-baseline": "bun run scripts/test/visual-baseline.ts",
-    "build:binary": "bun run scripts/build/compile.ts",
-    "test:backend": "bun run scripts/test/run.ts --scope backend",
-    "test:integration": "bun run scripts/test/run.ts --scope integration",
-    "test-db:down": "docker compose -f docker-compose.test.yml down",
-    "dashboard:dev": "bun run --cwd dashboard dev",
-    "test-db:check": "bun run scripts/test/check-db.ts",
-    "dashboard:test": "bun run scripts/test/run.ts --scope dashboard",
-    "test-db:reset": "docker compose -f docker-compose.test.yml down -v && docker compose -f docker-compose.test.yml up -d --wait postgres-test",
-    "dashboard:build": "bun run --cwd dashboard build",
-    "dashboard:typecheck": "bun run --cwd dashboard typecheck",
-    "check:migrations": "bun run scripts/test/check-migrations.ts"
-  },
-  "dependencies": {
-    "drizzle-orm": "0.45.2",
-    "elysia": "2.0.0-beta.16",
-    "enquirer": "^2.4.1",
-    "ioredis": "6.0.0",
-    "jose": "^6.2.12",
-    "jsonrepair": "^3.15.0",
-    "pg": "8.23.0",
-    "pino": "^10.3.1",
-    "pino-pretty": "^13.1.3",
-    "protobufjs": "^7.6.5",
-    "socks-proxy-agent": "^10.1.0",
-    "systray2": "2.1.4",
-    "typebox": "1.3.27",
-    "ws": "8.22.0"
-  },
-  "devDependencies": {
-    "@types/pg": "8.11.10",
-    "@types/ws": "8.18.1",
-    "bun-types": "1.4.0",
-    "concurrently": "^10.0.5",
-    "drizzle-kit": "^0.31.11",
-    "typescript": "5.9.2"
-  }
-}
+{
+  "name": "cartethyia",
+  "version": "2.6.5",
+  "description": "A powerful, self-hosted reverse proxy for OpenAI- and Anthropic-compatible AI clients, powered by native Bun and Elysia. It serves both protocols from a single server, routes every model across any configured provider, and translates requests and responses in real time — streaming and prompt caching included. Underneath it runs a full delivery engine: per-account health tracking, quota and cooldown enforcement, automatic failover across pools and providers, usage accounting, and telemetry. Point OpenCode, Cline, Claude Code, Codex, or any compatible client at Cartethyia and it keeps its native protocol while the gateway absorbs every provider-specific difference behind it. Runs on embedded SQLite or external PostgreSQL with optional Redis coordination, ships as a global CLI with a system-tray console, and needs only Bun 1.4+ alongside Node.js/npm.",
+  "license": "GPL-3.0-only",
+  "type": "module",
+  "packageManager": "bun@1.4.0",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/risunCodex/Cartethyia.git"
+  },
+  "homepage": "https://github.com/risunCodex/Cartethyia#readme",
+  "bugs": {
+    "url": "https://github.com/risunCodex/Cartethyia/issues"
+  },
+  "keywords": [
+    "ai",
+    "gateway",
+    "llm",
+    "bun",
+    "self-hosted"
+  ],
+  "bin": {
+    "cartethyia": "scripts/cli/bin/cartethyia.cjs",
+    "carte": "scripts/cli/bin/carte.cjs"
+  },
+  "files": [
+    "scripts/cli/",
+    "src/",
+    "migrations/",
+    "dist/dashboard/",
+    ".env.example",
+    "README.md"
+  ],
+  "workspaces": [
+    "dashboard"
+  ],
+  "engines": {
+    "bun": ">=1.4.0"
+  },
+  "scripts": {
+    "dev": "bun run dev:stack",
+    "dev:watch": "bun run scripts/dev/supervisor.ts",
+    "test": "bun run scripts/test/run.ts",
+    "build": "bun run scripts/build/build.ts",
+    "prepack": "bun run dashboard:build",
+    "setup": "bun run scripts/cli/setup.ts",
+    "start": "bun run scripts/cli/main.ts start",
+    "doctor": "bun run scripts/cli/doctor.ts",
+    "probe:egress": "bun run scripts/tools/probe-egress.ts",
+    "restart": "bun run scripts/build/restart.ts",
+    "build:aot": "bun run scripts/build/aot.ts",
+    "dev:stack": "concurrently -n BACKEND,DASHBOARD -c cyan,magenta --kill-others-on-fail \"bun run dev:backend\" \"bun run dashboard:dev\"",
+    "typecheck": "tsc --noEmit",
+    "test:watch": "bun run scripts/test/run.ts --watch",
+    "test-db:up": "docker compose -f docker-compose.test.yml up -d --wait postgres-test",
+    "dev:backend": "bun run --hot src/main.ts",
+    "dev:console-auth": "bun run scripts/test/console-session.ts",
+    "dev:visual-baseline": "bun run scripts/test/visual-baseline.ts",
+    "build:binary": "bun run scripts/build/compile.ts",
+    "test:backend": "bun run scripts/test/run.ts --scope backend",
+    "test:integration": "bun run scripts/test/run.ts --scope integration",
+    "test-db:down": "docker compose -f docker-compose.test.yml down",
+    "dashboard:dev": "bun run --cwd dashboard dev",
+    "test-db:check": "bun run scripts/test/check-db.ts",
+    "dashboard:test": "bun run scripts/test/run.ts --scope dashboard",
+    "test-db:reset": "docker compose -f docker-compose.test.yml down -v && docker compose -f docker-compose.test.yml up -d --wait postgres-test",
+    "dashboard:build": "bun run --cwd dashboard build",
+    "dashboard:typecheck": "bun run --cwd dashboard typecheck",
+    "check:migrations": "bun run scripts/test/check-migrations.ts"
+  },
+  "dependencies": {
+    "drizzle-orm": "0.45.2",
+    "elysia": "2.0.0-beta.16",
+    "enquirer": "^2.4.1",
+    "ioredis": "6.0.0",
+    "jose": "^6.2.12",
+    "jsonrepair": "^3.15.0",
+    "pg": "8.23.0",
+    "pino": "^10.3.1",
+    "pino-pretty": "^13.1.3",
+    "protobufjs": "^7.6.5",
+    "socks-proxy-agent": "^10.1.0",
+    "systray2": "2.1.4",
+    "typebox": "1.3.27",
+    "ws": "8.22.0"
+  },
+  "devDependencies": {
+    "@types/pg": "8.11.10",
+    "@types/ws": "8.18.1",
+    "bun-types": "1.4.0",
+    "concurrently": "^10.0.5",
+    "drizzle-kit": "^0.31.11",
+    "typescript": "5.9.2"
+  }
+}

package/src/boot/app.ts

@@ -269,7 +269,17 @@ export function createGatewayApp(deps: GatewayAppDeps) {
   const INFLIGHT_BACKSTOP_INTERVAL_MS = 30_000;
   const INFLIGHT_BACKSTOP_GRACE_MS = 30_000;
   const INFLIGHT_BACKSTOP_HARD_GRACE_MS = 120_000;
-  if (deps.mode === "production") deps.scheduledTasks?.register({
+  // Registered in every mode, not just production. The gauge is the operator's
+  // concurrency signal, and a cancelled request that never ran a finalizer
+  // leaves its flight behind — gating the only thing that reaps those on
+  // production made a permanently wrong count depend on how the gateway was
+  // started. The registry exists in every mode, so there is no mode where
+  // running this is wrong.
+  // `scheduledTasks` is declared on the production shape only, so it is
+  // narrowed rather than reached through the union — a shell composition
+  // without the registry simply skips the backstop, exactly as before.
+  const backstopTasks = "scheduledTasks" in deps ? deps.scheduledTasks : undefined;
+  backstopTasks?.register({
     name: "inflight-backstop",
     intervalMs: INFLIGHT_BACKSTOP_INTERVAL_MS,
     run: () => {

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

@@ -509,6 +509,13 @@ export interface ProviderAccountResponse {
   credentialKind: CredentialKind;
   credentialMode?: CredentialMode;
   tokenExpiresAt?: string;
+  /**
+   * Who issued this credential, read from the JWT `iss` claim at registration
+   * time. Present only for static JWTs, and only when the payload carried the
+   * claim: the secret is never readable after registration, so this is the one
+   * record of where the token came from.
+   */
+  jwtIssuer?: string;
   status: string;
   /** Live routing admission count for this request path; absent when unavailable. */
   inflight?: number;

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

@@ -59,11 +59,26 @@ function parseExpiry(value: unknown): Date | undefined {
   }
   return undefined;
 }
-function parseJwtCredential(raw: string): { readonly expiresAt?: Date } | undefined {
+/**
+ * Reads the two claims a static token's own payload can answer: when it stops
+ * working, and who issued it.
+ *
+ * The issuer is kept because a static JWT carries no other hint of its origin:
+ * an operator who pasted one from the wrong surface has nothing in the console
+ * to check it against, and several issuers accept the same token shape.
+ */
+function parseJwtCredential(
+  raw: string,
+): { readonly expiresAt?: Date; readonly issuer?: string } | undefined {
   const payload = decodeJwtPayload(raw.trim());
   if (payload === undefined) return undefined;
   const expiresAt = parseExpiry(payload["exp"]);
-  return expiresAt === undefined ? {} : { expiresAt };
+  const iss = payload["iss"];
+  const issuer = typeof iss === "string" && iss.trim().length > 0 ? iss.trim() : undefined;
+  return {
+    ...(expiresAt === undefined ? {} : { expiresAt }),
+    ...(issuer === undefined ? {} : { issuer }),
+  };
 }
 
 /**
@@ -1124,6 +1139,7 @@ export class DrizzleProviderCatalogStore implements ProviderCatalogStore {
         const requestedCredentialMode: CredentialMode = request.credentialMode ?? "auto";
         let effectiveCredentialMode: CredentialMode = requestedCredentialMode;
         let tokenExpiresAt: Date | undefined;
+        let jwtIssuer: string | undefined;
         let oauthRefreshCiphertext: Buffer | undefined;
         let oauthExpiresAt: Date | undefined;
         let oauthHasRefreshToken = false;
@@ -1160,6 +1176,9 @@ export class DrizzleProviderCatalogStore implements ProviderCatalogStore {
             effectiveCredentialKind = "api_key";
             effectiveCredentialMode = "jwt";
             tokenExpiresAt = parsedJwt.expiresAt;
+            // Kept alongside the expiry: the console shows who issued a static
+            // token, and the secret itself is never readable after this point.
+            jwtIssuer = parsedJwt.issuer;
           } else if (request.credentialKind === "oauth" || hasRefreshToken) {
             effectiveSecret = parts.accessToken;
             effectiveCredentialKind = hasRefreshToken ? "oauth" : "api_key";
@@ -1185,6 +1204,7 @@ export class DrizzleProviderCatalogStore implements ProviderCatalogStore {
           ...(request.authState ?? {}),
           credentialMode: effectiveCredentialMode,
           ...(tokenExpiresAt === undefined ? {} : { tokenExpiresAt: tokenExpiresAt.toISOString() }),
+          ...(jwtIssuer === undefined ? {} : { jwtIssuer }),
         };
         const nextSortIndex =
           sortIndexOverride ?? (await this.nextAccountSortIndex(tx, providerId, tenantId));
@@ -1409,6 +1429,8 @@ export class DrizzleProviderCatalogStore implements ProviderCatalogStore {
         : undefined;
     const tokenExpiresAt =
       typeof authState?.["tokenExpiresAt"] === "string" ? authState["tokenExpiresAt"] : undefined;
+    const jwtIssuer =
+      typeof authState?.["jwtIssuer"] === "string" ? authState["jwtIssuer"] : undefined;
     return {
       id: row.id,
       providerId: row.providerId,
@@ -1417,6 +1439,7 @@ export class DrizzleProviderCatalogStore implements ProviderCatalogStore {
       credentialKind: row.credentialKind,
       ...(credentialMode === undefined ? {} : { credentialMode }),
       ...(tokenExpiresAt === undefined ? {} : { tokenExpiresAt }),
+      ...(jwtIssuer === undefined ? {} : { jwtIssuer }),
       status: row.status,
       ...(usage.inflight === undefined ? {} : { inflight: usage.inflight }),
       usageToday: usage.today,

package/src/main.ts

@@ -52,6 +52,12 @@ const app = boot
       modelStrikes: boot.deps.modelStrikes,
       maxBodyBytes: resolveMaxBodyBytes(),
       scheduledTasks: boot.deps.scheduledTasks,
+      // The sampler is constructed in the dependency graph and already
+      // sampling on its own interval; without passing it here the app cannot
+      // bind its in-flight source, so `cartethyia_process_in_flight` stays
+      // pinned at the constructor's `() => 0` placeholder forever while the
+      // registry it was meant to read reports real flights.
+      runtimeMetricsSampler: boot.deps.runtimeMetricsSampler,
       shutdownCoordinator: boot.shutdownCoordinator,
       // Signal-free graceful stop for platforms where a catchable signal cannot
       // be delivered (Windows). Off unless `CARTETHYIA_DRAIN_TOKEN` is set.

package/src/observability/event-query.ts

@@ -215,7 +215,8 @@ export function eventRecordFromRow(row: {
  */
 function isPostgres(db: CartethyiaDatabase): boolean {
   const dialect = (db as unknown as { dialect?: { constructor?: { name?: string } } }).dialect;
-  return dialect?.constructor?.name === "PostgresDialect";
+  const name = dialect?.constructor?.name;
+  return name === "PostgresDialect" || name === "PgDialect";
 }
 
 /**
@@ -508,9 +509,14 @@ export async function chartTelemetry(
   bucketSeconds: number,
 ): Promise<ChartBucket[]> {
   const width = Math.max(1, Math.trunc(bucketSeconds));
+  // Inlined (not parameterised) so Postgres sees the SELECT and GROUP BY
+  // expressions as identical: parameter numbers differing between the two
+  // clauses made Postgres treat them as unrelated expressions referencing
+  // `created_at`, which 42803-rejected the query.
+  const w = sql.raw(String(width));
   const bucketExpression = isPostgres(db)
-    ? sql<number>`(cast(extract(epoch from ${telemetryEvents.createdAt}) as bigint) / ${width}) * ${width}`
-    : sql<number>`(cast(strftime('%s', ${telemetryEvents.createdAt}) as integer) / ${width}) * ${width}`;
+    ? sql<number>`(cast(extract(epoch from ${telemetryEvents.createdAt}) as bigint) / ${w}) * ${w}`
+    : sql<number>`(cast(strftime('%s', ${telemetryEvents.createdAt}) as integer) / ${w}) * ${w}`;
   const rows = await db
     .select({
       at: bucketExpression,
@@ -521,8 +527,7 @@ export async function chartTelemetry(
     })
     .from(telemetryEvents)
     .where(scoped(tenantId, since))
-    .groupBy(bucketExpression)
-    .orderBy(bucketExpression);
+    .groupBy(bucketExpression);
   return rows.map((row) => ({
     at: numeric(row.at) * 1000,
     requests: numeric(row.requests),