feat(memory): support custom remote embedding endpoints (#9622)

changelog.d/features/9620-cache-read-write-logs.md

@@ -1,2 +0,0 @@
-- Show cache-read and cache-write token counts in request log rows and details when providers
-  report them.

changelog.d/features/9622-memory-embedding-custom-endpoint.md

@@ -0,0 +1 @@
+- feat(memory): support custom OpenAI-compatible endpoints for Memory embeddings (#9622)

docs/frameworks/MEMORY.md

@@ -161,13 +161,15 @@ The `memory_vec_meta` table (migration `073_memory_vec.sql`) stores:
 
 ## Settings extension
 
-Seven new fields were added to `MemorySettingsExtended` (plan 21, D9) in
+Nine embedding and vector fields are available in `MemorySettingsExtended` in
 `src/shared/schemas/memory.ts`, persisted via `src/lib/db/settings.ts`:
 
 | Field                    | Type                                               | Default  | Description                                      |
 | ------------------------ | -------------------------------------------------- | -------- | ------------------------------------------------ |
 | `embeddingSource`        | `"remote" \| "static" \| "transformers" \| "auto"` | `"auto"` | Which embedding source to use                    |
 | `embeddingProviderModel` | `string \| null`                                   | `null`   | Provider/model in `provider/model` format        |
+| `customBaseUrl`          | `string \| null`                                   | `null`   | Memory-only OpenAI-compatible endpoint base URL  |
+| `customModelId`          | `string \| null`                                   | `null`   | Model ID sent to the custom endpoint             |
 | `transformersEnabled`    | `boolean`                                          | `false`  | Opt-in for Transformers.js (MiniLM, ~400MB)      |
 | `staticEnabled`          | `boolean`                                          | `false`  | Opt-in for static potion-base-8M local model     |
 | `rerankEnabled`          | `boolean`                                          | `false`  | Enable reranking step (adds +200-500ms/req)      |
@@ -176,6 +178,14 @@ Seven new fields were added to `MemorySettingsExtended` (plan 21, D9) in
 
 These are exposed via `GET /PUT /api/settings/memory` (schema `MemorySettingsExtendedSchema`).
 
+For the `remote` source, Memory also accepts the optional `customBaseUrl` and
+`customModelId` settings. Together they select an OpenAI-compatible `/embeddings`
+endpoint and model without changing the global embedding registry. The endpoint is
+normalized before use and checked by the provider outbound URL policy: HTTP(S) is
+required, embedded credentials and query strings are rejected, and cloud-metadata
+addresses remain blocked. Empty values preserve the selected registry provider. Errors
+returned to the dashboard are sanitized and endpoint credentials are never logged.
+
 > **TODO (D20):** Scope `global` (sharing memories across all API keys) is not
 > implemented in this release. It requires schema changes and a global retrieval
 > path. Track separately.

src/app/(dashboard)/dashboard/memory/components/CustomEmbeddingEndpointFields.tsx

@@ -0,0 +1,104 @@
+"use client";
+
+import { useState } from "react";
+import { useTranslations } from "next-intl";
+import type { MemorySettingsExtended } from "@/shared/schemas/memory";
+
+type Props = {
+  settings: MemorySettingsExtended;
+  onSave: (updates: Partial<MemorySettingsExtended>) => Promise<boolean>;
+  saving?: boolean;
+};
+
+function normalizeBaseUrl(value: string): string | null {
+  const trimmed = value.trim();
+  if (!trimmed) return null;
+  try {
+    const url = new URL(trimmed);
+    if (
+      (url.protocol !== "http:" && url.protocol !== "https:") ||
+      url.username ||
+      url.password ||
+      url.search ||
+      url.hash
+    ) {
+      return null;
+    }
+    return trimmed.replace(/\/+$/, "");
+  } catch {
+    return null;
+  }
+}
+
+export default function CustomEmbeddingEndpointFields({ settings, onSave, saving }: Props) {
+  const t = useTranslations("memory");
+  const [baseUrl, setBaseUrl] = useState(settings.customBaseUrl ?? "");
+  const [modelId, setModelId] = useState(settings.customModelId ?? "");
+  const [error, setError] = useState<string | null>(null);
+
+  const save = async () => {
+    const normalizedBaseUrl = normalizeBaseUrl(baseUrl);
+    const normalizedModelId = modelId.trim() || null;
+    if ((baseUrl.trim() || normalizedModelId) && (!normalizedBaseUrl || !normalizedModelId)) {
+      setError(t("embedding.customEndpointInvalid"));
+      return;
+    }
+    setError(null);
+    const saved = await onSave({
+      customBaseUrl: normalizedBaseUrl,
+      customModelId: normalizedModelId,
+    });
+    if (!saved) {
+      setError(t("embedding.customEndpointSaveFailed"));
+      return;
+    }
+    setBaseUrl(normalizedBaseUrl ?? "");
+    setModelId(normalizedModelId ?? "");
+  };
+
+  return (
+    <div className="mt-4 pt-4 border-t border-border/60 space-y-3">
+      <div>
+        <label className="block text-sm font-medium text-text-main mb-1">
+          {t("embedding.customBaseUrlLabel")}
+        </label>
+        <input
+          value={baseUrl}
+          onChange={(event) => setBaseUrl(event.target.value)}
+          placeholder="http://localhost:8000/v1"
+          disabled={saving}
+          data-testid="embedding-custom-base-url"
+          className="w-full px-3 py-2 rounded-lg bg-background border border-border text-sm"
+        />
+      </div>
+      <div>
+        <label className="block text-sm font-medium text-text-main mb-1">
+          {t("embedding.customModelIdLabel")}
+        </label>
+        <input
+          value={modelId}
+          onChange={(event) => setModelId(event.target.value)}
+          placeholder="my-embedding-model"
+          disabled={saving}
+          data-testid="embedding-custom-model-id"
+          className="w-full px-3 py-2 rounded-lg bg-background border border-border text-sm"
+        />
+      </div>
+      <p className="text-xs text-text-muted">{t("embedding.customEndpointHelp")}</p>
+      {error && (
+        <p role="alert" className="text-xs text-red-400">
+          {error}
+        </p>
+      )}
+      <button
+        type="button"
+        onClick={() => void save()}
+        disabled={saving}
+        data-testid="embedding-custom-save"
+        className="px-3 py-2 rounded-lg bg-violet-500 text-white text-sm disabled:opacity-50"
+      >
+        {t("embedding.customEndpointSave")}
+      </button>
+    </div>
+  );
+}

src/app/(dashboard)/dashboard/memory/components/EmbeddingSourceSelector.tsx

@@ -3,6 +3,7 @@
 import { useTranslations } from "next-intl";
 import type { MemorySettingsExtended } from "@/shared/schemas/memory";
 import type { EmbeddingProviderListing } from "@/lib/memory/embedding/types";
+import CustomEmbeddingEndpointFields from "./CustomEmbeddingEndpointFields";
 
 interface Props {
   settings: MemorySettingsExtended;
@@ -101,10 +102,11 @@ export default function EmbeddingSourceSelector({ settings, providers, onSave, s
                   <option key={m.id} value={m.id}>
                     {m.name} ({m.dimensions ? `${m.dimensions}d` : "?"})
                   </option>
-                )),
+                ))
               )}
             </select>
           )}
+          <CustomEmbeddingEndpointFields settings={settings} onSave={onSave} saving={saving} />
         </div>
       )}
 

src/i18n/messages/en.json

@@ -4142,6 +4142,12 @@
       "providerModelLabel": "Provider / Model",
       "noRemoteProviders": "No providers with configured API key",
       "selectProviderModel": "Select a model",
+      "customBaseUrlLabel": "Custom base URL (optional)",
+      "customModelIdLabel": "Custom model ID (optional)",
+      "customEndpointHelp": "Use an OpenAI-compatible /embeddings endpoint. Local endpoints are allowed; cloud metadata targets are blocked.",
+      "customEndpointSave": "Apply custom endpoint",
+      "customEndpointInvalid": "Enter both a valid HTTP(S) base URL and a model ID, or leave both fields empty.",
+      "customEndpointSaveFailed": "The custom embedding endpoint could not be saved.",
       "staticEnabledLabel": "Enable Static Potion",
       "staticEnabledDesc": "Download and use potion-base-8M model locally",
       "transformersEnabledLabel": "Enable Transformers.js",

src/i18n/messages/pt-BR.json

@@ -4142,6 +4142,12 @@
       "providerModelLabel": "Provider / Modelo",
       "noRemoteProviders": "Nenhum provider com chave configurada",
       "selectProviderModel": "Selecione um modelo",
+      "customBaseUrlLabel": "URL base personalizada (opcional)",
+      "customModelIdLabel": "ID do modelo personalizado (opcional)",
+      "customEndpointHelp": "Use um endpoint /embeddings compatível com OpenAI. Endpoints locais são permitidos; alvos de metadados de nuvem são bloqueados.",
+      "customEndpointSave": "Aplicar endpoint personalizado",
+      "customEndpointInvalid": "Informe uma URL base HTTP(S) válida e um ID de modelo, ou deixe os dois campos vazios.",
+      "customEndpointSaveFailed": "Não foi possível salvar o endpoint de embedding personalizado.",
       "staticEnabledLabel": "Habilitar Static Potion",
       "staticEnabledDesc": "Baixa e usa o modelo potion-base-8M localmente",
       "transformersEnabledLabel": "Habilitar Transformers.js",

src/lib/embeddings/service.ts

@@ -50,6 +50,8 @@ export interface EmbeddingHandlerOptions {
   apiKeyId?: string | null;
   apiKeyName?: string | null;
   connectionId?: string | null;
+  resolvedProvider?: EmbeddingProvider | null;
+  resolvedModel?: string | null;
 }
 
 export async function createEmbeddingResponse(
@@ -151,7 +153,13 @@ export async function createEmbeddingResponse(
     log.error("EMBED", `Failed to load provider_nodes for embeddings: ${err}`);
   }
 
-  const { provider, model: resolvedModel } = parseEmbeddingModel(body.model, dynamicProviders);
+  const parsedModel = options.resolvedProvider
+    ? {
+        provider: options.resolvedProvider.id,
+        model: options.resolvedModel ?? body.model,
+      }
+    : parseEmbeddingModel(body.model, dynamicProviders);
+  const { provider, model: resolvedModel } = parsedModel;
   if (!provider) {
     return errorResponse(
       HTTP_STATUS.BAD_REQUEST,
@@ -160,7 +168,10 @@ export async function createEmbeddingResponse(
   }
 
   let providerConfig: EmbeddingProvider | null =
-    dynamicProviders.find((dp) => dp.id === provider) || getEmbeddingProvider(provider) || null;
+    options.resolvedProvider ||
+    dynamicProviders.find((dp) => dp.id === provider) ||
+    getEmbeddingProvider(provider) ||
+    null;
   let credentialsProviderId = provider;
 
   if (!providerConfig) {

src/lib/memory/embedding/customProvider.ts

@@ -0,0 +1,62 @@
+import type { EmbeddingProvider } from "@omniroute/open-sse/config/embeddingRegistry.ts";
+import {
+  parseAndValidateNonMetadataUrl,
+  parseAndValidatePublicUrl,
+} from "@/shared/network/outboundUrlGuard";
+import { getProviderValidationGuard } from "@/shared/network/outboundUrlGuardPolicy";
+
+type CustomEmbeddingSettings = {
+  customBaseUrl?: string | null;
+  customModelId?: string | null;
+};
+
+export type ResolvedMemoryCustomEmbeddingProvider = {
+  provider: EmbeddingProvider;
+  model: string;
+  identity: string;
+};
+
+export class MemoryCustomEmbeddingConfigError extends Error {
+  constructor() {
+    super("Custom embedding endpoint is invalid or blocked");
+    this.name = "MemoryCustomEmbeddingConfigError";
+  }
+}
+
+function validateEndpoint(rawBaseUrl: string): URL {
+  const guard = getProviderValidationGuard();
+  if (guard === "public-only") return parseAndValidatePublicUrl(rawBaseUrl);
+  return parseAndValidateNonMetadataUrl(rawBaseUrl);
+}
+
+function toEmbeddingsUrl(url: URL): string {
+  if (url.search || url.hash) throw new MemoryCustomEmbeddingConfigError();
+  const normalized = url.toString().replace(/\/+$/, "");
+  return normalized.endsWith("/embeddings") ? normalized : `${normalized}/embeddings`;
+}
+
+export function resolveMemoryCustomEmbeddingProvider(
+  settings: CustomEmbeddingSettings
+): ResolvedMemoryCustomEmbeddingProvider | null {
+  const rawBaseUrl = settings.customBaseUrl?.trim() ?? "";
+  const model = settings.customModelId?.trim() ?? "";
+  if (!rawBaseUrl && !model) return null;
+  if (!rawBaseUrl || !model) throw new MemoryCustomEmbeddingConfigError();
+
+  try {
+    const baseUrl = toEmbeddingsUrl(validateEndpoint(rawBaseUrl));
+    return {
+      provider: {
+        id: "memory-custom",
+        baseUrl,
+        authType: "none",
+        authHeader: "none",
+        models: [],
+      },
+      model,
+      identity: `${baseUrl}|${model}`,
+    };
+  } catch {
+    throw new MemoryCustomEmbeddingConfigError();
+  }
+}

src/lib/memory/embedding/index.ts

@@ -17,6 +17,7 @@ import { embedRemote } from "./remote";
 import { embedStatic } from "./staticPotion";
 import { embedTransformers } from "./transformersLocal";
 import { buildCacheKey, get as cacheGet, set as cacheSet } from "./cache";
+import { resolveMemoryCustomEmbeddingProvider } from "./customProvider";
 
 const STATIC_MODEL = process.env.MEMORY_STATIC_MODEL || "minishlab/potion-base-8M";
 const TRANSFORMERS_MODEL = process.env.MEMORY_TRANSFORMERS_MODEL || "Xenova/all-MiniLM-L6-v2";
@@ -66,6 +67,22 @@ function remoteResolution(model: string, reasonPrefix: string): EmbeddingResolut
   };
 }
 
+function customRemoteResolution(settings: MemorySettingsExtended): EmbeddingResolution | null {
+  const customBaseUrl = settings.customBaseUrl?.trim() ?? "";
+  const customModelId = settings.customModelId?.trim() ?? "";
+  if (!customBaseUrl && !customModelId) return null;
+  if (!customBaseUrl || !customModelId) return noSource("custom embedding endpoint is incomplete");
+  const identity = `${customBaseUrl.replace(/\/+$/, "")}|${customModelId}`;
+  return {
+    source: "remote",
+    model: `memory-custom/${customModelId}`,
+    dimensions: null,
+    identity,
+    signature: makeSignature("remote", identity, null),
+    reason: "custom remote provider configured (dim=unknown, will probe at embed time)",
+  };
+}
+
 /**
  * Resolve which embedding source is active for the given settings (D4).
  * Pure: no heavy I/O. Provider key check done via synchronous registry lookup.
@@ -73,6 +90,11 @@ function remoteResolution(model: string, reasonPrefix: string): EmbeddingResolut
 export function resolveEmbeddingSource(settings: MemorySettingsExtended): EmbeddingResolution {
   const source = settings.embeddingSource ?? "auto";
 
+  const customResolution = customRemoteResolution(settings);
+  if (customResolution && (source === "remote" || source === "auto")) {
+    return customResolution;
+  }
+
   if (source === "remote") {
     // Explicit remote — check if the configured model has a key
     const model = settings.embeddingProviderModel ?? null;
@@ -194,7 +216,12 @@ export async function embed(
     };
   }
 
-  const cacheKey = buildCacheKey(resolution.source, resolution.model, resolution.dimensions, text);
+  const cacheKey = buildCacheKey(
+    resolution.source,
+    resolution.identity ?? resolution.model,
+    resolution.dimensions,
+    text
+  );
 
   const cached = cacheGet(cacheKey);
   if (cached) {
@@ -211,7 +238,21 @@ export async function embed(
   let result: EmbeddingResult | EmbeddingError;
 
   if (resolution.source === "remote") {
-    result = await embedRemote(text, resolution.model ?? "");
+    let customProvider;
+    try {
+      customProvider = resolveMemoryCustomEmbeddingProvider(settings);
+    } catch (error: unknown) {
+      return {
+        source: "remote",
+        model: resolution.model,
+        reason: "request_failed",
+        message:
+          error instanceof Error
+            ? error.message
+            : "Custom embedding endpoint is invalid or blocked",
+      };
+    }
+    result = await embedRemote(text, resolution.model ?? "", customProvider);
   } else if (resolution.source === "static") {
     result = await embedStatic(text);
   } else {

src/lib/memory/embedding/remote.ts

@@ -1,16 +1,26 @@
 import { sanitizeErrorMessage } from "@omniroute/open-sse/utils/error.ts";
 import { createEmbeddingResponse } from "@/lib/embeddings/service";
 import type { EmbeddingResult, EmbeddingError } from "./types";
+import type { ResolvedMemoryCustomEmbeddingProvider } from "./customProvider";
 
 export async function embedRemote(
   text: string,
-  model: string
+  model: string,
+  customProvider: ResolvedMemoryCustomEmbeddingProvider | null = null
 ): Promise<EmbeddingResult | EmbeddingError> {
   const t0 = Date.now();
 
   let resp: Response;
   try {
-    resp = await createEmbeddingResponse({ model, input: text });
+    resp = await createEmbeddingResponse(
+      { model, input: text },
+      customProvider
+        ? {
+            resolvedProvider: customProvider.provider,
+            resolvedModel: customProvider.model,
+          }
+        : undefined
+    );
   } catch (err: unknown) {
     // Network-level errors (ECONNREFUSED, AbortError, etc.)
     const isTimeout =
@@ -71,7 +81,9 @@ export async function embedRemote(
         source: "remote",
         model,
         reason: "request_failed",
-        message: sanitizeErrorMessage("Unexpected embedding response shape: missing data[0].embedding"),
+        message: sanitizeErrorMessage(
+          "Unexpected embedding response shape: missing data[0].embedding"
+        ),
       };
     }
     const rawVec = data[0].embedding as number[];

src/lib/memory/embedding/types.ts

@@ -19,6 +19,8 @@ export interface EmbeddingResolution {
   dimensions: number | null;
   /** Assinatura única usada como chave do vectorStore para detectar troca de modelo. */
   signature: string; // ${source}:${model}:${dim}
+  /** Cache/signature identity when the same model ID can exist at multiple custom endpoints. */
+  identity?: string;
   /** Motivo da escolha (UI exibe no Engine status). */
   reason: string; // e.g. "provider openai com key configurada"
 }
@@ -35,6 +37,7 @@ export interface EmbeddingResult {
 export interface EmbeddingError {
   source: "remote" | "static" | "transformers";
   model: string | null;
-  reason: "no_key" | "model_load_failed" | "request_failed" | "rate_limited" | "timeout" | "unknown";
+  reason:
+    "no_key" | "model_load_failed" | "request_failed" | "rate_limited" | "timeout" | "unknown";
   message: string; // ALWAYS via sanitizeErrorMessage()
 }

src/lib/memory/settings.ts

@@ -10,6 +10,8 @@ export interface MemorySettings {
   // Plan 21 — D9: new embedding / vector store fields
   embeddingSource: "remote" | "static" | "transformers" | "auto";
   embeddingProviderModel: string | null;
+  customBaseUrl: string | null;
+  customModelId: string | null;
   transformersEnabled: boolean;
   staticEnabled: boolean;
   rerankEnabled: boolean;
@@ -36,6 +38,8 @@ export const DEFAULT_MEMORY_SETTINGS: MemorySettings = {
   // Plan 21 — D9 defaults
   embeddingSource: "auto",
   embeddingProviderModel: null,
+  customBaseUrl: null,
+  customModelId: null,
   transformersEnabled: false,
   staticEnabled: false,
   rerankEnabled: false,
@@ -81,6 +85,17 @@ function normalizeNullableString(value: unknown, fallback: string | null): strin
   return typeof value === "string" && value.length > 0 ? value : fallback;
 }
 
+function normalizeCustomString(value: unknown): string | null {
+  if (typeof value !== "string") return null;
+  const normalized = value.trim();
+  return normalized.length > 0 ? normalized : null;
+}
+
+function normalizeCustomBaseUrl(value: unknown): string | null {
+  const normalized = normalizeCustomString(value);
+  return normalized ? normalized.replace(/\/+$/, "") : null;
+}
+
 export function normalizeMemorySettings(rawSettings: Record<string, unknown> = {}): MemorySettings {
   return {
     enabled: toBoolean(rawSettings.memoryEnabled, DEFAULT_MEMORY_SETTINGS.enabled),
@@ -104,6 +119,8 @@ export function normalizeMemorySettings(rawSettings: Record<string, unknown> = {
       rawSettings.memoryEmbeddingProviderModel,
       DEFAULT_MEMORY_SETTINGS.embeddingProviderModel
     ),
+    customBaseUrl: normalizeCustomBaseUrl(rawSettings.memoryEmbeddingCustomBaseUrl),
+    customModelId: normalizeCustomString(rawSettings.memoryEmbeddingCustomModelId),
     transformersEnabled: toBoolean(
       rawSettings.memoryTransformersEnabled,
       DEFAULT_MEMORY_SETTINGS.transformersEnabled
@@ -152,6 +169,10 @@ export function toMemorySettingsUpdates(
     updates.memoryEmbeddingSource = settings.embeddingSource;
   if (settings.embeddingProviderModel !== undefined)
     updates.memoryEmbeddingProviderModel = settings.embeddingProviderModel;
+  if (settings.customBaseUrl !== undefined)
+    updates.memoryEmbeddingCustomBaseUrl = settings.customBaseUrl;
+  if (settings.customModelId !== undefined)
+    updates.memoryEmbeddingCustomModelId = settings.customModelId;
   if (settings.transformersEnabled !== undefined)
     updates.memoryTransformersEnabled = settings.transformersEnabled;
   if (settings.staticEnabled !== undefined) updates.memoryStaticEnabled = settings.staticEnabled;

src/shared/components/RequestLoggerV2.tsx

@@ -95,7 +95,6 @@ const RequestLoggerV2 = forwardRef<RequestLoggerV2Handle, { initialSelectedId?:
   (props, ref) => {
     const { initialSelectedId } = props as any;
     const t = useTranslations("requestLogger");
-    const tCache = useTranslations("cache");
     const { emailsVisible } = useEmailPrivacyStore();
 
     // Get translated status filters
@@ -1515,30 +1514,6 @@ const RequestLoggerV2 = forwardRef<RequestLoggerV2Handle, { initialSelectedId?:
                                 <span className="text-emerald-700 dark:text-emerald-400">
                                   {log.tokens?.out?.toLocaleString() || 0}
                                 </span>
-                                {log.tokens?.cacheRead != null && log.tokens.cacheRead > 0 && (
-                                  <>
-                                    <span className="mx-1 text-border">|</span>
-                                    <span className="text-text-muted">CR:</span>{" "}
-                                    <span
-                                      className="text-sky-700 dark:text-sky-400"
-                                      title={tCache("cachedTokensCol")}
-                                    >
-                                      {log.tokens.cacheRead.toLocaleString()}
-                                    </span>
-                                  </>
-                                )}
-                                {log.tokens?.cacheWrite != null && log.tokens.cacheWrite > 0 && (
-                                  <>
-                                    <span className="mx-1 text-border">|</span>
-                                    <span className="text-text-muted">CW:</span>{" "}
-                                    <span
-                                      className="text-amber-700 dark:text-amber-400"
-                                      title={tCache("cacheCreation")}
-                                    >
-                                      {log.tokens.cacheWrite.toLocaleString()}
-                                    </span>
-                                  </>
-                                )}
                                 {log.tokens?.compressed != null && log.tokens.compressed > 0 && (
                                   <>
                                     <span className="mx-1 text-border">|</span>

src/shared/schemas/memory.ts

@@ -1,5 +1,34 @@
 import { z } from "zod";
 import { MemoryType } from "@/lib/memory/types";
+
+const optionalCustomEmbeddingValue = z.preprocess(
+  (value) => (typeof value === "string" && value.trim() === "" ? null : value),
+  z.string().trim().max(2048).nullable().optional()
+);
+
+const optionalCustomEmbeddingUrl = z.preprocess(
+  (value) => (typeof value === "string" && value.trim() === "" ? null : value),
+  z
+    .string()
+    .trim()
+    .max(2048)
+    .refine((value) => {
+      try {
+        const url = new URL(value);
+        return (
+          (url.protocol === "http:" || url.protocol === "https:") &&
+          !url.username &&
+          !url.password &&
+          !url.search &&
+          !url.hash
+        );
+      } catch {
+        return false;
+      }
+    }, "Custom embedding endpoint must be an HTTP(S) URL without credentials or query data")
+    .nullable()
+    .optional()
+);
 /** Schema estendido para PUT /api/settings/memory (D9). */
 export const MemorySettingsExtendedSchema = z
   .object({
@@ -12,6 +41,8 @@ export const MemorySettingsExtendedSchema = z
     // Campos novos (D9)
     embeddingSource: z.enum(["remote", "static", "transformers", "auto"]).optional(),
     embeddingProviderModel: z.string().nullable().optional(), // formato `provider/model`
+    customBaseUrl: optionalCustomEmbeddingUrl,
+    customModelId: optionalCustomEmbeddingValue,
     transformersEnabled: z.boolean().optional(),
     staticEnabled: z.boolean().optional(),
     rerankEnabled: z.boolean().optional(),

tests/unit/memory-embedding-custom-endpoint.test.ts

@@ -0,0 +1,156 @@
+import assert from "node:assert/strict";
+import { createServer } from "node:http";
+import { describe, it } from "node:test";
+import { MemorySettingsExtendedSchema } from "../../src/shared/schemas/memory.ts";
+import {
+  DEFAULT_MEMORY_SETTINGS,
+  normalizeMemorySettings,
+  toMemorySettingsUpdates,
+} from "../../src/lib/memory/settings.ts";
+import {
+  MemoryCustomEmbeddingConfigError,
+  resolveMemoryCustomEmbeddingProvider,
+} from "../../src/lib/memory/embedding/customProvider.ts";
+import { resolveEmbeddingSource } from "../../src/lib/memory/embedding/index.ts";
+import { EMBEDDING_PROVIDERS } from "@omniroute/open-sse/config/embeddingRegistry.ts";
+import { createEmbeddingResponse } from "../../src/lib/embeddings/service.ts";
+
+describe("Memory custom embedding endpoint", () => {
+  it("keeps the registry-backed behavior when custom fields are empty", () => {
+    const parsed = MemorySettingsExtendedSchema.parse({
+      customBaseUrl: "",
+      customModelId: "",
+    });
+    assert.equal(parsed.customBaseUrl, null);
+    assert.equal(parsed.customModelId, null);
+    assert.equal(resolveMemoryCustomEmbeddingProvider(parsed), null);
+  });
+
+  it("normalizes, persists, and resolves a Memory-only OpenAI-compatible provider", () => {
+    const settings = normalizeMemorySettings({
+      memoryEmbeddingCustomBaseUrl: " http://localhost:8000/v1/ ",
+      memoryEmbeddingCustomModelId: " SuperPauly/harrier-oss-v1-0.6b-gguf ",
+    });
+    assert.equal(settings.customBaseUrl, "http://localhost:8000/v1");
+    assert.equal(settings.customModelId, "SuperPauly/harrier-oss-v1-0.6b-gguf");
+
+    const resolved = resolveMemoryCustomEmbeddingProvider(settings);
+    assert.ok(resolved);
+    assert.equal(resolved.provider.id, "memory-custom");
+    assert.equal(resolved.provider.baseUrl, "http://localhost:8000/v1/embeddings");
+    assert.equal(resolved.provider.authType, "none");
+    assert.equal(resolved.model, "SuperPauly/harrier-oss-v1-0.6b-gguf");
+    assert.equal(EMBEDDING_PROVIDERS["memory-custom"], undefined);
+
+    const resolution = resolveEmbeddingSource({
+      embeddingSource: "remote",
+      embeddingProviderModel: null,
+      customBaseUrl: settings.customBaseUrl,
+      customModelId: settings.customModelId,
+    });
+    assert.equal(resolution.source, "remote");
+    assert.equal(resolution.model, "memory-custom/SuperPauly/harrier-oss-v1-0.6b-gguf");
+    assert.match(resolution.signature, /localhost:8000/);
+
+    assert.deepEqual(
+      toMemorySettingsUpdates({
+        customBaseUrl: settings.customBaseUrl,
+        customModelId: settings.customModelId,
+      }),
+      {
+        memoryEmbeddingCustomBaseUrl: "http://localhost:8000/v1",
+        memoryEmbeddingCustomModelId: "SuperPauly/harrier-oss-v1-0.6b-gguf",
+      }
+    );
+  });
+
+  it("preserves an endpoint that already ends in /embeddings", () => {
+    const resolved = resolveMemoryCustomEmbeddingProvider({
+      customBaseUrl: "https://embeddings.example.test/v1/embeddings/",
+      customModelId: "custom-model",
+    });
+    assert.equal(resolved?.provider.baseUrl, "https://embeddings.example.test/v1/embeddings");
+  });
+
+  it("rejects malformed, non-http, credential-bearing, query-bearing, and metadata URLs", () => {
+    const blocked = [
+      "not-a-url",
+      "file:///tmp/embeddings",
+      "https://user:secret@example.test/v1",
+      "https://example.test/v1?api_key=secret",
+      "http://169.254.169.254/latest/meta-data",
+    ];
+    for (const customBaseUrl of blocked) {
+      if (customBaseUrl !== "http://169.254.169.254/latest/meta-data") {
+        assert.equal(MemorySettingsExtendedSchema.safeParse({ customBaseUrl }).success, false);
+      }
+      assert.throws(
+        () =>
+          resolveMemoryCustomEmbeddingProvider({
+            customBaseUrl,
+            customModelId: "custom-model",
+          }),
+        (error: unknown) => {
+          assert.ok(error instanceof MemoryCustomEmbeddingConfigError);
+          assert.equal(error.message, "Custom embedding endpoint is invalid or blocked");
+          assert.equal(error.message.includes(customBaseUrl), false);
+          return true;
+        }
+      );
+    }
+  });
+
+  it("requires both custom fields and leaves defaults disabled", () => {
+    assert.equal(DEFAULT_MEMORY_SETTINGS.customBaseUrl, null);
+    assert.equal(DEFAULT_MEMORY_SETTINGS.customModelId, null);
+    assert.throws(
+      () =>
+        resolveMemoryCustomEmbeddingProvider({
+          customBaseUrl: "http://localhost:8000/v1",
+          customModelId: null,
+        }),
+      MemoryCustomEmbeddingConfigError
+    );
+  });
+
+  it("dispatches the custom model to a disposable OpenAI-compatible server", async () => {
+    let receivedPath = "";
+    let receivedBody: Record<string, unknown> | null = null;
+    const server = createServer((request, response) => {
+      receivedPath = request.url ?? "";
+      const chunks: Buffer[] = [];
+      request.on("data", (chunk: Buffer) => chunks.push(chunk));
+      request.on("end", () => {
+        receivedBody = JSON.parse(Buffer.concat(chunks).toString("utf8")) as Record<
+          string,
+          unknown
+        >;
+        response.writeHead(200, { "Content-Type": "application/json" });
+        response.end(JSON.stringify({ data: [{ embedding: [0.1, 0.2, 0.3] }] }));
+      });
+    });
+    await new Promise<void>((resolve) => server.listen(0, "127.0.0.1", resolve));
+
+    try {
+      const address = server.address();
+      assert.ok(address && typeof address === "object");
+      const custom = resolveMemoryCustomEmbeddingProvider({
+        customBaseUrl: `http://127.0.0.1:${address.port}/v1`,
+        customModelId: "custom-model",
+      });
+      assert.ok(custom);
+      const response = await createEmbeddingResponse(
+        { model: "memory-custom/custom-model", input: "hello" },
+        { resolvedProvider: custom.provider, resolvedModel: custom.model }
+      );
+      assert.equal(response.status, 200);
+      assert.equal(receivedPath, "/v1/embeddings");
+      assert.equal(receivedBody?.model, "custom-model");
+      assert.equal(EMBEDDING_PROVIDERS["memory-custom"], undefined);
+    } finally {
+      await new Promise<void>((resolve, reject) =>
+        server.close((error) => (error ? reject(error) : resolve()))
+      );
+    }
+  });
+});

tests/unit/ui/memory-embedding-custom-endpoint.test.tsx

@@ -0,0 +1,86 @@
+// @vitest-environment jsdom
+import React, { act } from "react";
+import { createRoot } from "react-dom/client";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+
+vi.mock("next-intl", () => ({
+  useTranslations: () => (key: string) => key,
+}));
+
+function changeInput(input: HTMLInputElement, value: string) {
+  const setter = Object.getOwnPropertyDescriptor(HTMLInputElement.prototype, "value")?.set;
+  setter?.call(input, value);
+  input.dispatchEvent(new Event("input", { bubbles: true }));
+}
+
+describe("Memory custom embedding endpoint controls", () => {
+  beforeEach(() => {
+    (
+      globalThis as typeof globalThis & { IS_REACT_ACT_ENVIRONMENT?: boolean }
+    ).IS_REACT_ACT_ENVIRONMENT = true;
+  });
+
+  afterEach(() => {
+    document.body.innerHTML = "";
+    vi.clearAllMocks();
+  });
+
+  it("validates locally and persists a complete custom endpoint override", async () => {
+    const { default: EmbeddingSourceSelector } =
+      await import("@/app/(dashboard)/dashboard/memory/components/EmbeddingSourceSelector");
+    const onSave = vi.fn().mockResolvedValue(true);
+    const container = document.createElement("div");
+    document.body.appendChild(container);
+    const root = createRoot(container);
+
+    await act(async () => {
+      root.render(
+        <EmbeddingSourceSelector
+          settings={{
+            embeddingSource: "remote",
+            embeddingProviderModel: null,
+            customBaseUrl: null,
+            customModelId: null,
+          }}
+          providers={[]}
+          onSave={onSave}
+        />
+      );
+    });
+
+    const baseUrl = container.querySelector(
+      "[data-testid='embedding-custom-base-url']"
+    ) as HTMLInputElement;
+    const modelId = container.querySelector(
+      "[data-testid='embedding-custom-model-id']"
+    ) as HTMLInputElement;
+    const save = container.querySelector(
+      "[data-testid='embedding-custom-save']"
+    ) as HTMLButtonElement;
+    expect(baseUrl).toBeTruthy();
+    expect(modelId).toBeTruthy();
+
+    await act(async () => {
+      changeInput(baseUrl, "file:///tmp/embeddings");
+    });
+    await act(async () => {
+      changeInput(modelId, "custom-model");
+    });
+    await act(async () => {
+      save.click();
+    });
+    expect(container.textContent).toContain("embedding.customEndpointInvalid");
+    expect(onSave).not.toHaveBeenCalled();
+
+    await act(async () => {
+      changeInput(baseUrl, "http://localhost:8000/v1/");
+    });
+    await act(async () => {
+      save.click();
+    });
+    expect(onSave).toHaveBeenCalledWith({
+      customBaseUrl: "http://localhost:8000/v1",
+      customModelId: "custom-model",
+    });
+  });
+});

tests/unit/ui/request-logger-cache-tokens.test.tsx

@@ -1,158 +0,0 @@
-// @vitest-environment jsdom
-import React, { act } from "react";
-import { createRoot, type Root } from "react-dom/client";
-import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
-
-vi.mock("next-intl", () => ({
-  useTranslations: (namespace?: string) => (key: string) =>
-    namespace === "cache"
-      ? ({ cachedTokensCol: "Cache Read", cacheCreation: "Cache Write" }[key] ?? key)
-      : key,
-}));
-
-vi.mock("next/navigation", () => ({
-  useRouter: () => ({ push: vi.fn(), replace: vi.fn(), refresh: vi.fn() }),
-}));
-
-vi.mock("@/store/emailPrivacyStore", () => ({
-  default: () => ({ emailsVisible: true }),
-}));
-
-const RequestLoggerV2 = (await import("@/shared/components/RequestLoggerV2")).default;
-const RequestLoggerDetail = (await import("@/shared/components/RequestLoggerDetail")).default;
-
-let container: HTMLElement;
-let root: Root;
-
-const populatedLog = {
-  id: "log-cache",
-  status: 200,
-  method: "POST",
-  path: "/v1/chat/completions",
-  model: "gpt-cache",
-  provider: "openai",
-  timestamp: "2026-08-10T12:00:00.000Z",
-  duration: 1_000,
-  tokens: {
-    in: 1_000,
-    out: 250,
-    cacheRead: 800,
-    cacheWrite: 120,
-    reasoning: 50,
-    compressed: 20,
-  },
-};
-
-const noop = () => {};
-
-async function render(component: React.ReactNode) {
-  await act(async () => {
-    root.render(component);
-  });
-}
-
-beforeEach(() => {
-  container = document.createElement("div");
-  document.body.appendChild(container);
-  root = createRoot(container);
-});
-
-afterEach(async () => {
-  await act(async () => {
-    root.unmount();
-  });
-  container.remove();
-  vi.unstubAllGlobals();
-});
-
-describe("request log cache token metrics (#9620)", () => {
-  it("renders cache read/write beside the existing row token metrics", async () => {
-    const emptyCacheLog = {
-      ...populatedLog,
-      id: "log-no-cache",
-      model: "gpt-no-cache",
-      timestamp: "2026-08-10T11:59:00.000Z",
-      tokens: { ...populatedLog.tokens, cacheRead: null, cacheWrite: 0 },
-    };
-    vi.stubGlobal(
-      "fetch",
-      vi.fn(async (input: RequestInfo | URL) => {
-        const url = String(input);
-        if (url.startsWith("/api/usage/call-logs")) {
-          return Response.json([populatedLog, emptyCacheLog]);
-        }
-        if (url.startsWith("/api/provider-nodes")) return Response.json({ nodes: [] });
-        if (url.startsWith("/api/logs/detail")) return Response.json({ enabled: false });
-        return Response.json({});
-      })
-    );
-
-    await render(<RequestLoggerV2 />);
-    await act(async () => {
-      await Promise.resolve();
-    });
-
-    const row = Array.from(container.querySelectorAll("tbody tr")).find((candidate) =>
-      candidate.textContent?.includes("gpt-cache")
-    );
-    expect(row?.textContent).toContain("TI: 1,000");
-    expect(row?.textContent).toContain("TO: 250");
-    expect(row?.textContent).toContain("CR: 800");
-    expect(row?.textContent).toContain("CW: 120");
-    expect(row?.textContent).toContain("↓20");
-
-    const emptyRow = Array.from(container.querySelectorAll("tbody tr")).find((candidate) =>
-      candidate.textContent?.includes("gpt-no-cache")
-    );
-    expect(emptyRow?.textContent).toContain("TI: 1,000");
-    expect(emptyRow?.textContent).toContain("TO: 250");
-    expect(emptyRow?.textContent).not.toContain("CR:");
-    expect(emptyRow?.textContent).not.toContain("CW:");
-  });
-
-  it("distinguishes cache read from cache write in the detail view", async () => {
-    await render(
-      <RequestLoggerDetail
-        log={populatedLog}
-        detail={populatedLog}
-        loading={false}
-        debugEnabled={false}
-        onClose={noop}
-        onCopy={async () => true}
-      />
-    );
-
-    const inputGroup = container.querySelector('[data-testid="token-group-input"]');
-    const outputGroup = container.querySelector('[data-testid="token-group-output"]');
-    expect(inputGroup?.textContent).toContain("Total In: 1,000");
-    expect(inputGroup?.textContent).toContain("Cache Read: 800");
-    expect(inputGroup?.textContent).toContain("Cache Write: 120");
-    expect(inputGroup?.textContent).toContain("Compressed:");
-    expect(outputGroup?.textContent).toContain("Total Out: 250");
-    expect(outputGroup?.textContent).toContain("Reasoning: 50");
-  });
-
-  it("handles historical null and zero cache values without inventing usage", async () => {
-    const emptyCacheLog = {
-      ...populatedLog,
-      id: "log-no-cache",
-      tokens: { ...populatedLog.tokens, cacheRead: null, cacheWrite: 0 },
-    };
-
-    await render(
-      <RequestLoggerDetail
-        log={emptyCacheLog}
-        detail={emptyCacheLog}
-        loading={false}
-        debugEnabled={false}
-        onClose={noop}
-        onCopy={async () => true}
-      />
-    );
-
-    const inputGroup = container.querySelector('[data-testid="token-group-input"]');
-    expect(inputGroup?.textContent).toContain("Cache Read: N/A");
-    expect(inputGroup?.textContent).toContain("Cache Write: 0");
-    expect(inputGroup?.textContent).toContain("Total In: 1,000");
-  });
-});