@vellumai/assistant 0.10.4-dev.202607010918.ba1a70c → 0.10.4-dev.202607011113.fffc936

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vellumai/assistant",
-  "version": "0.10.4-dev.202607010918.ba1a70c",
+  "version": "0.10.4-dev.202607011113.fffc936",
   "license": "MIT",
   "type": "module",
   "exports": {

package/src/daemon/lifecycle.ts

@@ -31,6 +31,7 @@ import {
 } from "../persistence/conversation-crud.js";
 import { getDb } from "../persistence/db-connection.js";
 import { initializeDb } from "../persistence/db-init.js";
+import { startEmbeddingRuntimeManager } from "../persistence/embeddings/embedding-runtime-manager.js";
 import { startConsentRefresh } from "../platform/consent-cache.js";
 import { syncWorkspaceIdentityToPlatform } from "../platform/sync-identity.js";
 import { runMemoryStartup } from "../plugins/defaults/memory/startup.js";
@@ -879,30 +880,7 @@ export async function runDaemon(): Promise<void> {
     log.warn({ err }, "Assistant symlink installation failed — continuing");
   }
 
-  // Download embedding runtime in background (non-blocking).
-  // If download fails, local embeddings gracefully fall back to cloud backends.
-  void (async () => {
-    try {
-      const { EmbeddingRuntimeManager } =
-        await import("../persistence/embeddings/embedding-runtime-manager.js");
-      const runtimeManager = new EmbeddingRuntimeManager();
-      if (!runtimeManager.isReady()) {
-        log.info("Downloading embedding runtime in background...");
-        await runtimeManager.ensureInstalled();
-        // Reset the sticky local-backend failure flag so auto mode retries
-        // local embeddings without evicting a worker that may already be live.
-        const { resetLocalEmbeddingFailureState } =
-          await import("../persistence/embeddings/embedding-backend.js");
-        resetLocalEmbeddingFailureState();
-        log.info("Embedding runtime download complete");
-      }
-    } catch (err) {
-      log.warn(
-        { err },
-        "Embedding runtime download failed — local embeddings will use cloud fallback",
-      );
-    }
-  })();
+  startEmbeddingRuntimeManager();
 
   startWorkspaceHeartbeatService();
 

package/src/persistence/embeddings/embedding-runtime-manager.ts

@@ -571,3 +571,34 @@ export class EmbeddingRuntimeManager {
     }
   }
 }
+
+/**
+ * Download the local embedding runtime in the background (non-blocking).
+ *
+ * Pre-warms the runtime (bun binary + ONNX worker scripts) so the first local
+ * embed doesn't pay the download cost inline. A no-op when the runtime is
+ * already installed. Failures are swallowed with a warning: local embeddings
+ * fall back to cloud backends, so a failed download must never block or crash
+ * daemon startup. Returns immediately; the download proceeds on its own.
+ */
+export function startEmbeddingRuntimeManager(): void {
+  void (async () => {
+    try {
+      const runtimeManager = new EmbeddingRuntimeManager();
+      if (runtimeManager.isReady()) return;
+      log.info("Downloading embedding runtime in background...");
+      await runtimeManager.ensureInstalled();
+      // Reset the sticky local-backend failure flag so auto mode retries
+      // local embeddings without evicting a worker that may already be live.
+      const { resetLocalEmbeddingFailureState } =
+        await import("./embedding-backend.js");
+      resetLocalEmbeddingFailureState();
+      log.info("Embedding runtime download complete");
+    } catch (err) {
+      log.warn(
+        { err },
+        "Embedding runtime download failed — local embeddings will use cloud fallback",
+      );
+    }
+  })();
+}

package/src/runtime/migrations/origin-mode.ts

@@ -7,15 +7,15 @@
  *   - `hasManagedProxyPrereqs()` — true when the daemon has the credentials
  *     it needs to act as a managed-proxy client, i.e. this is a managed
  *     deployment.
- *   - `getDaemonRuntimeMode()` — `"docker"` vs `"bare-metal"`, identifying
- *     where the daemon process is running.
+ *   - `getIsContainerized()` — true when the daemon runs inside a container
+ *     (`IS_CONTAINERIZED`), identifying where the daemon process is running.
  *
  * Folding both into a single helper keeps callers from repeating the
  * combination logic.
  */
 
+import { getIsContainerized } from "../../config/env-registry.js";
 import { hasManagedProxyPrereqs } from "../../providers/platform-proxy/context.js";
-import { getDaemonRuntimeMode } from "../runtime-mode.js";
 
 export type VBundleOriginMode =
   | "managed"
@@ -26,14 +26,14 @@ export type VBundleOriginMode =
  * Returns the origin mode for the current daemon.
  *
  * Managed-proxy prereqs win first (a managed deployment is always
- * "managed" regardless of where the daemon process runs); otherwise docker
- * → "self-hosted-remote", bare-metal → "self-hosted-local".
+ * "managed" regardless of where the daemon process runs); otherwise
+ * containerized → "self-hosted-remote", bare-metal → "self-hosted-local".
  */
 export async function getOriginMode(): Promise<VBundleOriginMode> {
   if (await hasManagedProxyPrereqs()) {
     return "managed";
   }
-  if (getDaemonRuntimeMode() === "docker") {
+  if (getIsContainerized()) {
     return "self-hosted-remote";
   }
   return "self-hosted-local";

package/src/runtime/__tests__/runtime-mode.test.ts

@@ -1,62 +0,0 @@
-/**
- * Tests for `getDaemonRuntimeMode()` — ensures the helper reflects
- * `IS_CONTAINERIZED` environment state via the existing truthy-check
- * semantics shared with `getIsContainerized()`.
- */
-
-import { afterEach, beforeEach, describe, expect, test } from "bun:test";
-
-import { getDaemonRuntimeMode } from "../runtime-mode.js";
-
-describe("getDaemonRuntimeMode", () => {
-  let savedIsContainerized: string | undefined;
-
-  beforeEach(() => {
-    savedIsContainerized = process.env.IS_CONTAINERIZED;
-  });
-
-  afterEach(() => {
-    if (savedIsContainerized === undefined) {
-      delete process.env.IS_CONTAINERIZED;
-    } else {
-      process.env.IS_CONTAINERIZED = savedIsContainerized;
-    }
-  });
-
-  test("returns 'docker' when IS_CONTAINERIZED=true", () => {
-    process.env.IS_CONTAINERIZED = "true";
-    expect(getDaemonRuntimeMode()).toBe("docker");
-  });
-
-  test("returns 'docker' when IS_CONTAINERIZED=1", () => {
-    // Matches the existing truthy-check in getIsContainerized(), which
-    // treats "1" as equivalent to "true" for boolean env flags.
-    process.env.IS_CONTAINERIZED = "1";
-    expect(getDaemonRuntimeMode()).toBe("docker");
-  });
-
-  test("returns 'bare-metal' when IS_CONTAINERIZED is unset", () => {
-    delete process.env.IS_CONTAINERIZED;
-    expect(getDaemonRuntimeMode()).toBe("bare-metal");
-  });
-
-  test("returns 'bare-metal' when IS_CONTAINERIZED=false", () => {
-    process.env.IS_CONTAINERIZED = "false";
-    expect(getDaemonRuntimeMode()).toBe("bare-metal");
-  });
-
-  test("returns 'bare-metal' when IS_CONTAINERIZED=0", () => {
-    process.env.IS_CONTAINERIZED = "0";
-    expect(getDaemonRuntimeMode()).toBe("bare-metal");
-  });
-
-  test("returns 'bare-metal' when IS_CONTAINERIZED is an empty string", () => {
-    process.env.IS_CONTAINERIZED = "";
-    expect(getDaemonRuntimeMode()).toBe("bare-metal");
-  });
-
-  test("returns 'bare-metal' for arbitrary non-truthy values", () => {
-    process.env.IS_CONTAINERIZED = "yes";
-    expect(getDaemonRuntimeMode()).toBe("bare-metal");
-  });
-});

package/src/runtime/runtime-mode.ts

@@ -1,38 +0,0 @@
-/**
- * Daemon runtime-mode detection.
- *
- * Exposes a single well-named helper `getDaemonRuntimeMode()` that returns
- * `"docker"` when the daemon is running inside a container and
- * `"bare-metal"` otherwise.
- *
- * Under the hood this delegates to `getIsContainerized()` from the env
- * registry, which accepts the standard truthy values for `IS_CONTAINERIZED`
- * (`"true"` or `"1"`). Keeping the check in the registry avoids duplicating
- * the env-parsing semantics across modules.
- *
- * The mode-named API (rather than a boolean) exists to make downstream
- * switch/branch code read naturally — e.g. `if (mode === "docker") { ... }`
- * is clearer than `if (isContainerized) { ... }` when the behavior depends
- * on the specific deployment shape, and it leaves room for additional
- * runtime modes in the future without renaming every callsite.
- *
- * The `DaemonRuntimeMode` type is declared here, alongside the resolver
- * that produces it. `"docker"` describes a daemon running inside a
- * container; `"bare-metal"` describes a daemon running directly on the
- * host.
- */
-
-import { getIsContainerized } from "../config/env-registry.js";
-
-export type DaemonRuntimeMode = "bare-metal" | "docker";
-
-/**
- * Returns the deployment mode the daemon is currently running under.
- *
- * - `"docker"` when `IS_CONTAINERIZED` is set to a truthy value
- *   (`"true"` or `"1"`).
- * - `"bare-metal"` otherwise.
- */
-export function getDaemonRuntimeMode(): DaemonRuntimeMode {
-  return getIsContainerized() ? "docker" : "bare-metal";
-}