veryfront 0.1.630 → 0.1.631

package/esm/src/provider/local/local-engine.js

@@ -15,6 +15,7 @@ import { createError, toError } from "../../errors/veryfront-error.js";
 import { importTransformers } from "../../platform/compat/opaque-deps.js";
 import { DEFAULT_LOCAL_MODEL, resolveLocalModel } from "./model-catalog.js";
 import { throwIfLocalAIDisabled } from "./env.js";
+import { createPipelineCache } from "./pipeline-cache.js";
 const logger = serverLogger.component("local-llm");
 /** Default maximum new tokens for local model generation */
 const DEFAULT_MAX_NEW_TOKENS = 512;
@@ -82,10 +83,6 @@ export function buildPipeOptions(options, transformers, tokenizer, streamer) {
     }
     return pipeOptions;
 }
-/** Cached pipeline instances keyed by HuggingFace model ID */
-const pipelineCache = new Map();
-/** Whether a model is currently being loaded (prevents concurrent loads) */
-const loadingLocks = new Map();
 let transformersModule = null;
 /**
  * Lazily import @huggingface/transformers.
@@ -117,21 +114,12 @@ export async function getTransformers() {
     return mod;
 }
 /**
- * Load a text-generation pipeline for the given model.
- * Returns a cached pipeline if already loaded.
+ * Bounded, dedup-aware cache of text-generation pipelines keyed by HuggingFace
+ * model id. Only loads a model on a cold cache miss; concurrent loads of the
+ * same model share a single promise.
  */
-async function loadPipeline(modelInfo) {
-    const cacheKey = modelInfo.hfId;
-    // Return cached pipeline
-    const cached = pipelineCache.get(cacheKey);
-    if (cached)
-        return cached;
-    // Wait for existing load if in progress
-    const existingLock = loadingLocks.get(cacheKey);
-    if (existingLock)
-        return existingLock;
-    // Start loading
-    const loadPromise = (async () => {
+const textGenerationPipelines = createPipelineCache(async (modelInfo) => {
+    try {
         const transformers = await getTransformers();
         logger.info(`Loading local model: ${modelInfo.hfId} (${modelInfo.dtype}, ~${modelInfo.sizeMB}MB)...`);
         const pipe = (await transformers.pipeline("text-generation", modelInfo.hfId, {
@@ -139,16 +127,9 @@ async function loadPipeline(modelInfo) {
             device: "cpu",
         }));
         logger.info(`Model loaded: ${modelInfo.hfId}`);
-        pipelineCache.set(cacheKey, pipe);
-        loadingLocks.delete(cacheKey);
         return pipe;
-    })();
-    loadingLocks.set(cacheKey, loadPromise);
-    try {
-        return await loadPromise;
     }
     catch (error) {
-        loadingLocks.delete(cacheKey);
         // Convert ONNX / native-addon errors to no_ai_available so they propagate
         // correctly through the chat handler (503) instead of being swallowed as
         // in-band SSE errors inside a 200 response stream.
@@ -166,6 +147,13 @@ async function loadPipeline(modelInfo) {
         }
         throw error;
     }
+});
+/**
+ * Load a text-generation pipeline for the given model.
+ * Returns a cached pipeline if already loaded.
+ */
+function loadPipeline(modelInfo) {
+    return textGenerationPipelines.load(modelInfo.hfId, modelInfo);
 }
 /**
  * Eagerly verify that the local AI runtime (@huggingface/transformers + ONNX)
@@ -262,5 +250,5 @@ export async function preloadModel(modelId) {
  */
 export function isModelLoaded(modelId) {
     const modelInfo = resolveLocalModel(modelId);
-    return pipelineCache.has(modelInfo.hfId);
+    return textGenerationPipelines.has(modelInfo.hfId);
 }

package/esm/src/provider/local/pipeline-cache.d.ts

@@ -0,0 +1,27 @@
+/**
+ * Shared pipeline cache for local Transformers.js engines.
+ *
+ * Both the text-generation engine (`local-engine.ts`) and the embedding engine
+ * (`local-embedding-engine.ts`) need to lazily load a pipeline per HuggingFace
+ * model id, cache it, and deduplicate concurrent loads of the same model. This
+ * helper captures that shared behavior and bounds the cache with an LRU so a
+ * long-running process that touches many models cannot grow memory without
+ * limit (loaded models are large).
+ *
+ * @module provider/local
+ */
+/** A loaded pipeline cache with concurrency-safe, deduplicated loading. */
+export interface PipelineCache<P, M> {
+    /** Load (or return cached) pipeline for the given model id. */
+    load(cacheKey: string, modelInfo: M): Promise<P>;
+    /** Whether a pipeline for the given model id is currently loaded. */
+    has(cacheKey: string): boolean;
+}
+/**
+ * Create a bounded, dedup-aware pipeline cache.
+ *
+ * @param loadFn Loads the underlying pipeline for a model. Only invoked on a
+ *   cold cache miss; concurrent loads of the same key share a single promise.
+ */
+export declare function createPipelineCache<P, M>(loadFn: (modelInfo: M) => Promise<P>): PipelineCache<P, M>;
+//# sourceMappingURL=pipeline-cache.d.ts.map

package/esm/src/provider/local/pipeline-cache.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"pipeline-cache.d.ts","sourceRoot":"","sources":["../../../../src/src/provider/local/pipeline-cache.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAUH,2EAA2E;AAC3E,MAAM,WAAW,aAAa,CAAC,CAAC,EAAE,CAAC;IACjC,+DAA+D;IAC/D,IAAI,CAAC,QAAQ,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IACjD,qEAAqE;IACrE,GAAG,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC;CAChC;AAED;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAAC,CAAC,EAAE,CAAC,EACtC,MAAM,EAAE,CAAC,SAAS,EAAE,CAAC,KAAK,OAAO,CAAC,CAAC,CAAC,GACnC,aAAa,CAAC,CAAC,EAAE,CAAC,CAAC,CAwCrB"}

package/esm/src/provider/local/pipeline-cache.js

@@ -0,0 +1,61 @@
+/**
+ * Shared pipeline cache for local Transformers.js engines.
+ *
+ * Both the text-generation engine (`local-engine.ts`) and the embedding engine
+ * (`local-embedding-engine.ts`) need to lazily load a pipeline per HuggingFace
+ * model id, cache it, and deduplicate concurrent loads of the same model. This
+ * helper captures that shared behavior and bounds the cache with an LRU so a
+ * long-running process that touches many models cannot grow memory without
+ * limit (loaded models are large).
+ *
+ * @module provider/local
+ */
+import { LRUCache } from "../../utils/lru-wrapper.js";
+/**
+ * Maximum number of loaded pipelines kept in memory at once. Loaded models are
+ * large (hundreds of MB each), so the cap is intentionally small.
+ */
+const PIPELINE_CACHE_MAX_ENTRIES = 4;
+/**
+ * Create a bounded, dedup-aware pipeline cache.
+ *
+ * @param loadFn Loads the underlying pipeline for a model. Only invoked on a
+ *   cold cache miss; concurrent loads of the same key share a single promise.
+ */
+export function createPipelineCache(loadFn) {
+    const pipelineCache = new LRUCache({
+        maxEntries: PIPELINE_CACHE_MAX_ENTRIES,
+    });
+    // Whether a model is currently being loaded (prevents concurrent loads).
+    const loadingLocks = new Map();
+    return {
+        async load(cacheKey, modelInfo) {
+            // Return cached pipeline.
+            const cached = pipelineCache.get(cacheKey);
+            if (cached)
+                return cached;
+            // Wait for existing load if in progress.
+            const existingLock = loadingLocks.get(cacheKey);
+            if (existingLock)
+                return existingLock;
+            // Start loading.
+            const loadPromise = (async () => {
+                const pipe = await loadFn(modelInfo);
+                pipelineCache.set(cacheKey, pipe);
+                loadingLocks.delete(cacheKey);
+                return pipe;
+            })();
+            loadingLocks.set(cacheKey, loadPromise);
+            try {
+                return await loadPromise;
+            }
+            catch (error) {
+                loadingLocks.delete(cacheKey);
+                throw error;
+            }
+        },
+        has(cacheKey) {
+            return pipelineCache.has(cacheKey);
+        },
+    };
+}

package/esm/src/server/handlers/dev/framework-candidates.generated.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"framework-candidates.generated.d.ts","sourceRoot":"","sources":["../../../../../src/src/server/handlers/dev/framework-candidates.generated.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,eAAO,MAAM,oBAAoB,EAAE,SAAS,MAAM,EAg/KjD,CAAC"}
+{"version":3,"file":"framework-candidates.generated.d.ts","sourceRoot":"","sources":["../../../../../src/src/server/handlers/dev/framework-candidates.generated.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,eAAO,MAAM,oBAAoB,EAAE,SAAS,MAAM,EAi/KjD,CAAC"}

package/esm/src/server/handlers/dev/framework-candidates.generated.js

@@ -5516,6 +5516,7 @@ export const FRAMEWORK_CANDIDATES = [
     "veryfront/chat",
     "veryfront/errors/error-registry.ts",
     "veryfront/platform/compat/runtime.ts",
+    "veryfront/react/primitives/index.ts",
     "veryfront/schemas/_test-setup.ts",
     "veryfront/security/client/html-sanitizer.ts",
     "veryfront/testing/assert",

package/esm/src/server/handlers/request/internal-agents-list.handler.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"internal-agents-list.handler.d.ts","sourceRoot":"","sources":["../../../../../src/src/server/handlers/request/internal-agents-list.handler.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAC;AAClD,OAAO,KAAK,EAAE,cAAc,EAAE,eAAe,EAAmB,aAAa,EAAE,MAAM,aAAa,CAAC;AACnG,OAAO,EAKL,KAAK,yBAAyB,EAC/B,MAAM,oCAAoC,CAAC;AAa5C,qBAAa,yBAA0B,SAAQ,WAAW;IAS5C,OAAO,CAAC,QAAQ,CAAC,IAAI;IARjC,QAAQ,EAAE,eAAe,CAMvB;gBAE2B,IAAI,GAAE,yBAAoD;IAIjF,MAAM,CAAC,GAAG,EAAE,OAAO,EAAE,GAAG,EAAE,cAAc,GAAG,OAAO,CAAC,aAAa,CAAC;CAkExE"}
+{"version":3,"file":"internal-agents-list.handler.d.ts","sourceRoot":"","sources":["../../../../../src/src/server/handlers/request/internal-agents-list.handler.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAC;AAClD,OAAO,KAAK,EAAE,cAAc,EAAE,eAAe,EAAmB,aAAa,EAAE,MAAM,aAAa,CAAC;AACnG,OAAO,EAML,KAAK,yBAAyB,EAC/B,MAAM,oCAAoC,CAAC;AAa5C,qBAAa,yBAA0B,SAAQ,WAAW;IAS5C,OAAO,CAAC,QAAQ,CAAC,IAAI;IARjC,QAAQ,EAAE,eAAe,CAMvB;gBAE2B,IAAI,GAAE,yBAAoD;IAIjF,MAAM,CAAC,GAAG,EAAE,OAAO,EAAE,GAAG,EAAE,cAAc,GAAG,OAAO,CAAC,aAAa,CAAC;CAmExE"}

package/esm/src/server/handlers/request/internal-agents-list.handler.js

@@ -30,6 +30,7 @@ export class InternalAgentsListHandler extends BaseHandler {
                 const payload = ControlPlaneAgentsListRequestSchema.parse(JSON.parse(rawBody));
                 const claims = await verifyControlPlaneRequest(req, ctx, rawBody, {
                     expectedSubject: payload.requestId,
+                    // Validated by `.parse()` above; narrow back to the union (object inference widens it to `string`).
                     expectedSurface: payload.surface,
                 });
                 if (payload.projectId !== claims.project_id ||

package/esm/src/transforms/mdx/esm-module-loader/module-writer.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"module-writer.d.ts","sourceRoot":"","sources":["../../../../../src/src/transforms/mdx/esm-module-loader/module-writer.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAyBH,OAAO,KAAK,EAAkB,SAAS,EAAE,MAAM,aAAa,CAAC;AAC7D,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAiDnD,wBAAsB,eAAe,CACnC,mBAAmB,EAAE,MAAM,EAC3B,OAAO,EAAE,gBAAgB,GACxB,OAAO,CAAC,SAAS,CAAC,CAkVpB"}
+{"version":3,"file":"module-writer.d.ts","sourceRoot":"","sources":["../../../../../src/src/transforms/mdx/esm-module-loader/module-writer.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAyBH,OAAO,KAAK,EAAkB,SAAS,EAAE,MAAM,aAAa,CAAC;AAC7D,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAiDnD,wBAAsB,eAAe,CACnC,mBAAmB,EAAE,MAAM,EAC3B,OAAO,EAAE,gBAAgB,GACxB,OAAO,CAAC,SAAS,CAAC,CA4VpB"}

package/esm/src/transforms/mdx/esm-module-loader/module-writer.js

@@ -184,7 +184,10 @@ export async function doLoadModuleESM(compiledProgramCode, context) {
                     compositeKey = `${namespaceKey}:${codeHash}`;
                     // Clean up orphaned module file if path changed
                     if (refreshedPath !== originalFilePath) {
-                        getLocalFs().remove(originalFilePath).catch(() => { });
+                        getLocalFs().remove(originalFilePath).catch((error) => logger.debug(`${LOG_PREFIX_MDX_LOADER} orphaned module file cleanup failed`, {
+                            originalFilePath,
+                            error,
+                        }));
                     }
                     // Verify bundles exist after re-fetch
                     const refreshedBundles = extractHttpBundlePaths(rewritten);
@@ -241,7 +244,10 @@ export async function doLoadModuleESM(compiledProgramCode, context) {
                 compositeKey = `${namespaceKey}:${codeHash}`;
                 // Clean up orphaned module file if path changed
                 if (refreshedPath !== originalFilePath) {
-                    getLocalFs().remove(originalFilePath).catch(() => { });
+                    getLocalFs().remove(originalFilePath).catch((error) => logger.debug(`${LOG_PREFIX_MDX_LOADER} orphaned module file cleanup failed`, {
+                        originalFilePath,
+                        error,
+                    }));
                 }
                 // Verify bundles now exist
                 const stillMissing = await findMissingFrameworkBundles(extractFrameworkBundlePaths(rewritten));

package/esm/src/transforms/pipeline/stages/ssr-vf-modules/transform.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"transform.d.ts","sourceRoot":"","sources":["../../../../../../src/src/transforms/pipeline/stages/ssr-vf-modules/transform.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,gBAAgB,EAAE,MAAM,mCAAmC,CAAC;AAarE,OAAO,EAOL,KAAK,gBAAgB,EAGtB,MAAM,gBAAgB,CAAC;AAKxB;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAExD;AAED;;;;;;;;;;GAUG;AACH,wBAAsB,oBAAoB,CACxC,WAAW,EAAE,MAAM,EACnB,YAAY,EAAE,MAAM,EACpB,EAAE,EAAE,UAAU,CAAC,OAAO,gBAAgB,CAAC,GACtC,OAAO,CAAC,MAAM,CAAC,CA2BjB;AAwED;;;;GAIG;AACH,wBAAgB,qBAAqB,CACnC,YAAY,EAAE,MAAM,EACpB,YAAY,EAAE,MAAM,EACpB,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GACtC,MAAM,GAAG,IAAI,CAKf;AAwMD;;;;;GAKG;AACH,wBAAsB,sBAAsB,CAC1C,OAAO,EAAE,MAAM,EACf,UAAU,EAAE,MAAM,EAClB,GAAG,EAAE,gBAAgB,EACrB,oBAAoB,UAAQ,EAC5B,KAAK,SAAI,GACR,OAAO,CAAC,MAAM,CAAC,CAsPjB;AAED;;;GAGG;AACH,wBAAsB,kCAAkC,CACtD,SAAS,EAAE,MAAM,EACjB,GAAG,EAAE,gBAAgB,GACpB,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CA4CxB;AAED;;;GAGG;AACH,wBAAsB,wBAAwB,CAC5C,OAAO,EAAE,MAAM,EACf,UAAU,EAAE,MAAM,EAClB,YAAY,EAAE,MAAM,EACpB,UAAU,EAAE,MAAM,EAClB,EAAE,EAAE,UAAU,CAAC,OAAO,gBAAgB,CAAC,GACtC,OAAO,CAAC,MAAM,CAAC,CAEjB"}
+{"version":3,"file":"transform.d.ts","sourceRoot":"","sources":["../../../../../../src/src/transforms/pipeline/stages/ssr-vf-modules/transform.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,gBAAgB,EAAE,MAAM,mCAAmC,CAAC;AAcrE,OAAO,EAOL,KAAK,gBAAgB,EAGtB,MAAM,gBAAgB,CAAC;AAKxB;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAExD;AAED;;;;;;;;;;GAUG;AACH,wBAAsB,oBAAoB,CACxC,WAAW,EAAE,MAAM,EACnB,YAAY,EAAE,MAAM,EACpB,EAAE,EAAE,UAAU,CAAC,OAAO,gBAAgB,CAAC,GACtC,OAAO,CAAC,MAAM,CAAC,CA2BjB;AA8ED;;;;GAIG;AACH,wBAAgB,qBAAqB,CACnC,YAAY,EAAE,MAAM,EACpB,YAAY,EAAE,MAAM,EACpB,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GACtC,MAAM,GAAG,IAAI,CAKf;AAwMD;;;;;GAKG;AACH,wBAAsB,sBAAsB,CAC1C,OAAO,EAAE,MAAM,EACf,UAAU,EAAE,MAAM,EAClB,GAAG,EAAE,gBAAgB,EACrB,oBAAoB,UAAQ,EAC5B,KAAK,SAAI,GACR,OAAO,CAAC,MAAM,CAAC,CAsPjB;AAED;;;GAGG;AACH,wBAAsB,kCAAkC,CACtD,SAAS,EAAE,MAAM,EACjB,GAAG,EAAE,gBAAgB,GACpB,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CA4CxB;AAED;;;GAGG;AACH,wBAAsB,wBAAwB,CAC5C,OAAO,EAAE,MAAM,EACf,UAAU,EAAE,MAAM,EAClB,YAAY,EAAE,MAAM,EACpB,UAAU,EAAE,MAAM,EAClB,EAAE,EAAE,UAAU,CAAC,OAAO,gBAAgB,CAAC,GACtC,OAAO,CAAC,MAAM,CAAC,CAEjB"}

package/esm/src/transforms/pipeline/stages/ssr-vf-modules/transform.js

@@ -10,6 +10,7 @@ import { rendererLogger as logger } from "../../../../utils/index.js";
 import { IMPORT_RESOLUTION_ERROR } from "../../../../errors/index.js";
 import { replaceSpecifiers } from "../../../esm/lexer.js";
 import { hashCodeHex } from "../../../../utils/hash-utils.js";
+import { LRUCache } from "../../../../utils/lru-wrapper.js";
 import { getHttpBundleCacheDir, getMdxEsmCacheDir } from "../../../../utils/cache-dir.js";
 import { cacheHttpImportsToLocal } from "../../../esm/http-cache.js";
 import { loadImportMap } from "../../../../modules/import-map/index.js";
@@ -69,7 +70,13 @@ export async function cacheTransformedCode(transformed, vfModulePath, fs) {
 // React, HTTP imports all rewritten). The fallback still reads
 // `frameworkFileCache` first, so when the main path has already produced
 // a high-quality entry the fallback prefers it.
-const fallbackTransformCache = new Map();
+// Bounded so the fallback's per-path output cannot grow memory without limit
+// in a long-running dev server. Entries are deterministic per resolved path and
+// safe to evict/recompute.
+const FALLBACK_TRANSFORM_CACHE_MAX_ENTRIES = 2000;
+const fallbackTransformCache = new LRUCache({
+    maxEntries: FALLBACK_TRANSFORM_CACHE_MAX_ENTRIES,
+});
 /**
  * Resolve a bare `react` / `react-dom` (or subpath) specifier to its esm.sh
  * URL for the given React version. Returns `null` for anything that is not a

package/esm/src/utils/version-constant.d.ts

@@ -1,3 +1,3 @@
 /** Shared version value. */
-export declare const VERSION = "0.1.630";
+export declare const VERSION = "0.1.631";
 //# sourceMappingURL=version-constant.d.ts.map

package/esm/src/utils/version-constant.js

@@ -1,4 +1,4 @@
 // Keep in sync with deno.json version.
 // scripts/release.ts updates this constant during releases.
 /** Shared version value. */
-export const VERSION = "0.1.630";
+export const VERSION = "0.1.631";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "veryfront",
-  "version": "0.1.630",
+  "version": "0.1.631",
   "description": "The simplest way to build AI-powered apps",
   "keywords": [
     "react",