Read embedding settings from runtime config

ts/packages/aiclient/src/embeddingProvider.ts

@@ -5,6 +5,7 @@ import { TextEmbeddingModel } from "./models.js";
 import { createEmbeddingModel } from "./openai.js";
 import { EnvVars } from "./apiTypes.js";
 import { createLocalEmbeddingModel } from "./localEmbedding.js";
+import { getRuntimeConfig } from "./runtimeConfig.js";
 import {
     createCopilotEmbeddingModel,
     DefaultCopilotEmbeddingModel,
@@ -23,60 +24,40 @@ export type EmbeddingProvider =
     | "copilot"
     | "none";
 
-// Flattened form of the config `embedding:` section.
-enum EmbeddingEnvVars {
-    PROVIDER = "TYPEAGENT_EMBEDDING_PROVIDER",
-    MODEL = "TYPEAGENT_EMBEDDING_MODEL",
-    CACHE_DIR = "TYPEAGENT_EMBEDDING_CACHE_DIR",
-    SIZE = "TYPEAGENT_EMBEDDING_SIZE",
-    MAX_BATCH_SIZE = "TYPEAGENT_EMBEDDING_MAX_BATCH_SIZE",
-}
-
 // Embedding sizes of well-known defaults, used when no size is configured.
 const LocalDefaultEmbeddingSize = 384; // Xenova/all-MiniLM-L6-v2
 const HostedDefaultEmbeddingSize = 1536; // ada-002 / text-embedding-3-small
 
-function readPositiveInt(name: string): number | undefined {
-    const raw = process.env[name]?.trim();
-    if (!raw) return undefined;
-    const n = Number(raw);
-    return Number.isInteger(n) && n > 0 ? n : undefined;
+// The `embedding:` section of the runtime config.
+function getEmbeddingConfig() {
+    return getRuntimeConfig().embedding;
 }
 
 /**
  * The embedding vector size for the configured provider. An explicit
- * `embedding.size` (`TYPEAGENT_EMBEDDING_SIZE`) always wins; otherwise the
- * default model's size is used (set `size` when using a non-default model) (384 for the local MiniLM model, 1536 for
- * hosted endpoints). Configuration only; never loads a model.
+ * `embedding.size` always wins; otherwise the default model's size is used:
+ * `LocalDefaultEmbeddingSize` for the local provider,
+ * `HostedDefaultEmbeddingSize` for hosted endpoints. Set `size` when using a
+ * non-default model. Configuration only; never loads a model.
  */
 export function getEmbeddingSize(): number {
-    const configured = readPositiveInt(EmbeddingEnvVars.SIZE);
+    const configured = getEmbeddingConfig()?.size;
     if (configured !== undefined) return configured;
     return getEmbeddingProvider() === "local"
         ? LocalDefaultEmbeddingSize
         : HostedDefaultEmbeddingSize;
 }
 
-function isEmbeddingProvider(value: string): value is EmbeddingProvider {
-    return (
-        value === "local" ||
-        value === "openai" ||
-        value === "azure" ||
-        value === "copilot" ||
-        value === "none"
-    );
-}
-
 /**
  * Determine the configured embedding provider using configuration only
  * (no network access, no model loading). An explicit
- * `TYPEAGENT_EMBEDDING_PROVIDER` always wins; otherwise the provider is
+ * `embedding.provider` always wins; otherwise the provider is
  * inferred from the presence of hosted embedding endpoints, defaulting to
  * "none" when nothing is configured.
  */
 export function getEmbeddingProvider(): EmbeddingProvider {
-    const explicit = process.env[EmbeddingEnvVars.PROVIDER]?.trim();
-    if (explicit && isEmbeddingProvider(explicit)) {
+    const explicit = getEmbeddingConfig()?.provider;
+    if (explicit !== undefined) {
         return explicit;
     }
     if (
@@ -113,37 +94,31 @@ export function tryCreateEmbeddingModel(
     dimensions?: number,
 ): TextEmbeddingModel | undefined {
     const provider = getEmbeddingProvider();
+    const config = getEmbeddingConfig();
     switch (provider) {
         case "none":
             return undefined;
         case "local":
             return createLocalEmbeddingModel({
-                model: process.env[EmbeddingEnvVars.MODEL]?.trim() || undefined,
-                cacheDir:
-                    process.env[EmbeddingEnvVars.CACHE_DIR]?.trim() ||
-                    undefined,
-                maxBatchSize: readPositiveInt(EmbeddingEnvVars.MAX_BATCH_SIZE),
+                model: config?.model,
+                cacheDir: config?.cacheDir,
+                maxBatchSize: config?.maxBatchSize,
             });
         case "copilot":
             return createCopilotEmbeddingModel(
-                process.env[EmbeddingEnvVars.MODEL]?.trim() ||
-                    DefaultCopilotEmbeddingModel,
+                config?.model ?? DefaultCopilotEmbeddingModel,
                 undefined,
                 undefined,
                 {
-                    dimensions:
-                        dimensions ?? readPositiveInt(EmbeddingEnvVars.SIZE),
-                    maxBatchSize: readPositiveInt(
-                        EmbeddingEnvVars.MAX_BATCH_SIZE,
-                    ),
+                    dimensions: dimensions ?? config?.size,
+                    maxBatchSize: config?.maxBatchSize,
                 },
             );
         default: {
-            dimensions ??= readPositiveInt(EmbeddingEnvVars.SIZE);
+            dimensions ??= config?.size;
             const options = {
-                modelName:
-                    process.env[EmbeddingEnvVars.MODEL]?.trim() || undefined,
-                maxBatchSize: readPositiveInt(EmbeddingEnvVars.MAX_BATCH_SIZE),
+                modelName: config?.model,
+                maxBatchSize: config?.maxBatchSize,
             };
             return endpoint !== undefined
                 ? createEmbeddingModel(endpoint, dimensions, options)

ts/packages/aiclient/src/openai.ts

@@ -877,6 +877,15 @@ export function createEmbeddingModel(
 
     // https://platform.openai.com/docs/api-reference/embeddings/create#embeddings-create-input
     const maxBatchSize = Math.min(options?.maxBatchSize ?? 2048, 2048);
+    // Trace config overrides of the pool's settings for troubleshooting.
+    if (
+        options?.modelName !== undefined ||
+        options?.maxBatchSize !== undefined
+    ) {
+        debugOpenAI(
+            `Embedding overrides for ${pool.modelKey}: model=${options.modelName ?? settings.modelName}, maxBatchSize=${maxBatchSize}`,
+        );
+    }
     const defaultParams: any =
         settings.provider === "azure"
             ? {}

ts/packages/config/src/runtime/build.ts

@@ -1006,7 +1006,8 @@ function buildEmbedding(
     };
 }
 
-// Pop a positive integer; malformed values are left in extra untouched.
+// Pop a positive integer; malformed values are left in extra untouched
+// and reported, so a typo (e.g. `size: -1`) does not silently fall back.
 function popPositiveInt(
     flat: Map<string, string>,
     key: string,
@@ -1015,6 +1016,9 @@ function popPositiveInt(
     if (raw === undefined) return undefined;
     const n = Number(raw);
     if (Number.isInteger(n) && n > 0) return n;
+    console.warn(
+        `${key}: expected a positive integer, got "${raw}"; ignoring.`,
+    );
     flat.set(key, raw);
     return undefined;
 }

ts/packages/knowPro/src/conversation.ts

@@ -32,11 +32,8 @@ export function createConversationSettings(
     // May be undefined when no embedding provider is configured (e.g. Copilot
     // self-host without a local embedder). Embedding-backed indexes then no-op
     // and search degrades to exact/alias/edit-distance matching.
-    if (embeddingModel === undefined) {
-        embeddingModel = tryCreateEmbeddingModel();
-        embeddingSize ??= getEmbeddingSize();
-    }
-    embeddingSize ??= 1536;
+    embeddingModel ??= tryCreateEmbeddingModel();
+    embeddingSize ??= getEmbeddingSize();
     if (embeddingModel === undefined && !embeddingUnavailableAnnounced) {
         embeddingUnavailableAnnounced = true;
         console.warn(