@infersec/conduit 1.16.0 → 1.17.0

package/dist/apiClient/index.d.ts

@@ -5,7 +5,7 @@ export interface APIClient {
     reportMachineMetadata: (payload: InferenceAgentMachineReportPayload) => Promise<void>;
     reportPromptMetrics: (payload: InferenceAgentLLMMetricsPayload) => Promise<void>;
 }
-export declare function createAPIClient({ apiURL, inferenceSourceID }: {
+export declare function createAPIClient({ apiKey, apiURL, inferenceSourceID }: {
     apiKey: string;
     apiURL: string;
     inferenceSourceID: ULID;

package/dist/cli.js

@@ -6,7 +6,7 @@ const __dirname = __pathDirname(__filename);
 
 import { parseArgs } from 'node:util';
 import 'node:crypto';
-import { a as asError, s as startInferenceAgent } from './start-DUnVCZr8.js';
+import { a as asError, s as startInferenceAgent } from './start-CCmUbFoS.js';
 import 'argon2';
 import 'node:child_process';
 import 'node:stream';

package/dist/index.js

@@ -5,7 +5,7 @@ const __filename = __fileURLToPath(import.meta.url);
 const __dirname = __pathDirname(__filename);
 
 import 'node:crypto';
-import { s as startInferenceAgent, a as asError } from './start-DUnVCZr8.js';
+import { s as startInferenceAgent, a as asError } from './start-CCmUbFoS.js';
 import 'argon2';
 import 'node:child_process';
 import 'node:stream';

package/dist/{start-DUnVCZr8.js → start-CCmUbFoS.js}

@@ -14791,6 +14791,11 @@ const LLMModelSchema = object({
         })
     ])
 });
+const QuantizationFileSchema = object({
+    filePath: string$1().min(1),
+    sizeBytes: number$1().int().nonnegative().nullable()
+});
+array(QuantizationFileSchema);
 
 const InferenceAgentConfigurationSchema = object({
     contextLength: number$1().int().positive().nullable(),
@@ -52620,6 +52625,19 @@ function requireUndici () {
 
 var undiciExports = requireUndici();
 
+function createFetchWithHeaders({ fetchFn = undiciExports.fetch, headers }) {
+    function fetchWithHeaders(url, options) {
+        const mergedHeaders = new Headers(options.headers);
+        for (const [key, value] of Object.entries(headers)) {
+            mergedHeaders.set(key, value);
+        }
+        return fetchFn(url, {
+            ...options,
+            headers: mergedHeaders
+        });
+    }
+    return fetchWithHeaders;
+}
 async function fetchByReference(options) {
     const { baseURL, body, fetch: fetchFn = undiciExports.fetch, parameters = {}, query = {}, reference, route, method } = options;
     if (!reference[route]) {
@@ -97901,11 +97919,17 @@ function implementAPIReference({ api, logger, mount, reference }) {
     }
 }
 
-function createAPIClient({ apiURL, inferenceSourceID }) {
+function createAPIClient({ apiKey, apiURL, inferenceSourceID }) {
+    const fetchWithAPIKey = createFetchWithHeaders({
+        headers: {
+            "x-api-key": apiKey
+        }
+    });
     return {
         getConduitConfiguration: async () => {
             const result = await fetchByReference({
                 baseURL: apiURL,
+                fetch: fetchWithAPIKey,
                 reference: API_SERVICE_CONDUIT_API_REFERENCE,
                 route: "/conduit/api/v1/source/:sourceID/configuration",
                 method: "GET",
@@ -97919,6 +97943,7 @@ function createAPIClient({ apiURL, inferenceSourceID }) {
             await fetchByReference({
                 baseURL: apiURL,
                 body: payload,
+                fetch: fetchWithAPIKey,
                 method: "POST",
                 parameters: {
                     sourceID: inferenceSourceID
@@ -97931,6 +97956,7 @@ function createAPIClient({ apiURL, inferenceSourceID }) {
             await fetchByReference({
                 baseURL: apiURL,
                 body: state,
+                fetch: fetchWithAPIKey,
                 method: "POST",
                 parameters: {
                     sourceID: inferenceSourceID
@@ -97943,6 +97969,7 @@ function createAPIClient({ apiURL, inferenceSourceID }) {
             await fetchByReference({
                 baseURL: apiURL,
                 body: payload,
+                fetch: fetchWithAPIKey,
                 method: "POST",
                 parameters: {
                     sourceID: inferenceSourceID
@@ -99576,6 +99603,24 @@ const ModelDownloadProgressSchema = object({
 const DOWNLOAD_PROGRESS_TIMEOUT = 60000;
 const DOWNLOAD_RETRY_ATTEMPTS_FULL = 3;
 const DOWNLOAD_RETRY_ATTEMPTS_RANGE = 10;
+function matchesQuantizationVariant({ filePath, variant }) {
+    if (!variant) {
+        return false;
+    }
+    const escapedVariant = variant.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+    const trailingBoundary = variant.includes("_") ? "[\\-./_]" : "[\\-./]";
+    const matcher = new RegExp(`(^|[\\-./_])${escapedVariant}(?=$|${trailingBoundary})`, "i");
+    const normalizedPath = filePath.replace(/\\/g, "/");
+    const segments = normalizedPath.split("/").filter(Boolean);
+    if (segments.length === 0) {
+        return false;
+    }
+    const filename = segments[segments.length - 1].replace(/\.gguf$/i, "");
+    if (matcher.test(filename)) {
+        return true;
+    }
+    return segments.slice(0, -1).some(segment => matcher.test(segment));
+}
 async function downloadModelViaHuggingFace({ format, huggingFaceToken, modelSlug: rawModelSlug, onProgress, progressFilePath, targetDirectory }) {
     // Sanitise model ID
     const [modelSlugWithRevision, variant = null] = rawModelSlug.split(":");
@@ -99607,7 +99652,10 @@ async function downloadModelViaHuggingFace({ format, huggingFaceToken, modelSlug
         }
         if (variant &&
             filePath.endsWith(".gguf") &&
-            filePath.includes(variant.toLowerCase()) === false) {
+            !matchesQuantizationVariant({
+                filePath: file.path,
+                variant
+            })) {
             console.log("Skipping due variant not being needed:", file.path, file.size);
             continue;
         }
@@ -118177,6 +118225,7 @@ async function startInferenceAgent({ configurationOverrides }) {
         agentEngineType: configuration.agentEngineType
     });
     const apiClient = createAPIClient({
+        apiKey: configuration.apiKey,
         apiURL: configuration.apiURL,
         inferenceSourceID: configuration.inferenceSourceID
     });

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@infersec/conduit",
   "description": "End user conduit agent for connecting local LLMs to the cloud.",
-  "version": "1.16.0",
+  "version": "1.17.0",
   "bin": {
     "infersec-conduit": "./dist/cli.js"
   },