@vellumai/assistant 0.5.16 → 0.6.1

package/src/cli/commands/memory.ts

@@ -7,7 +7,6 @@ import {
   getMemorySystemStatus,
   queryMemory,
   requestMemoryBackfill,
-  requestMemoryCleanup,
   requestMemoryRebuildIndex,
   requestReextract,
 } from "../../memory/admin.js";
@@ -36,7 +35,7 @@ Key concepts:
 Examples:
   $ assistant memory status
   $ assistant memory query "What is the project deadline?"
-  $ assistant memory backfill`,
+  $ assistant memory backfill`
   );
 
   memory
@@ -52,15 +51,13 @@ Fields shown:
                      degraded mode (e.g. missing embedding backend)
   embedding backend  The provider/model pair used for vector embeddings (or "none")
   segments           Total conversation segments indexed
-  items              Total distilled fact items stored
+  graph nodes        Total memory graph nodes stored
   summaries          Total compressed conversation summaries
   embeddings         Total vector embeddings computed
-  cleanup backlogs   Number of superseded items pending cleanup
-  cleanup throughput Number of cleanup operations completed in the last 24 hours
   jobs               Status of background jobs (backfill, cleanup, rebuild-index)
 
 Examples:
-  $ assistant memory status`,
+  $ assistant memory status`
     )
     .action(async () => {
       initializeDb();
@@ -74,15 +71,9 @@ Examples:
         log.info("Embedding backend: none");
       }
       log.info(`Segments: ${status.counts.segments.toLocaleString()}`);
-      log.info(`Items: ${status.counts.items.toLocaleString()}`);
+      log.info(`Graph nodes: ${status.counts.graphNodes.toLocaleString()}`);
       log.info(`Summaries: ${status.counts.summaries.toLocaleString()}`);
       log.info(`Embeddings: ${status.counts.embeddings.toLocaleString()}`);
-      log.info(
-        `Cleanup backlog (superseded items): ${status.cleanup.supersededBacklog.toLocaleString()}`,
-      );
-      log.info(
-        `Cleanup throughput 24h (superseded items): ${status.cleanup.supersededCompleted24h.toLocaleString()}`,
-      );
       log.info("Jobs:");
       for (const [key, value] of Object.entries(status.jobs)) {
         log.info(`  ${key}: ${value}`);
@@ -106,7 +97,7 @@ useful after bulk imports or if the incremental state has become inconsistent.
 
 Examples:
   $ assistant memory backfill
-  $ assistant memory backfill --force`,
+  $ assistant memory backfill --force`
     )
     .action((opts: { force?: boolean }) => {
       initializeDb();
@@ -114,40 +105,6 @@ Examples:
       log.info(`Queued backfill job: ${jobId}`);
     });
 
-  memory
-    .command("cleanup")
-    .description("Queue cleanup jobs for stale superseded items")
-    .option(
-      "--retention-ms <ms>",
-      "Optional retention threshold in milliseconds",
-    )
-    .addHelpText(
-      "after",
-      `
-Queues a background cleanup job to remove memory items that have been
-superseded by newer, corrected facts past the retention threshold.
-
-The optional --retention-ms flag sets the minimum age (in milliseconds) a
-record must have before it is eligible for cleanup. If omitted, the system
-default retention period is used.
-
-Examples:
-  $ assistant memory cleanup
-  $ assistant memory cleanup --retention-ms 86400000`,
-    )
-    .action((opts: { retentionMs?: string }) => {
-      initializeDb();
-      const retentionMs = opts.retentionMs
-        ? Number.parseInt(opts.retentionMs, 10)
-        : undefined;
-      const jobs = requestMemoryCleanup(
-        Number.isFinite(retentionMs) ? retentionMs : undefined,
-      );
-      log.info(
-        `Queued cleanup_stale_superseded_items job: ${jobs.staleSupersededItemsJobId}`,
-      );
-    });
-
   memory
     .command("cleanup-segments")
     .description("Remove short segments that waste retrieval budget")
@@ -156,7 +113,7 @@ Examples:
       "after",
       `
 Removes segments shorter than the minimum character threshold from both
-SQLite and Qdrant. Short fragments (e.g. "Collar Touched") burn embedding
+SQLite and Qdrant. Short fragments (e.g. "OK sounds good") burn embedding
 budget, retrieval slots, and injection tokens without adding value.
 
 New segments are already filtered at creation time. This command cleans up
@@ -164,19 +121,21 @@ existing short segments that were stored before the filter was added.
 
 Examples:
   $ assistant memory cleanup-segments
-  $ assistant memory cleanup-segments --dry-run`,
+  $ assistant memory cleanup-segments --dry-run`
     )
     .action(async (opts: { dryRun?: boolean }) => {
       initializeDb();
       const result = await cleanupShortSegments({ dryRun: opts.dryRun });
       if (opts.dryRun) {
         log.info(
-          `Dry run: ${result.dryRunCount} short segment(s) would be removed.`,
+          `Dry run: ${result.dryRunCount} short segment(s) would be removed.`
         );
       } else {
         log.info(`Removed ${result.removed} short segment(s).`);
         if (result.failed > 0) {
-          log.warn(`${result.failed} segment(s) skipped — Qdrant deletion failed. Re-run when Qdrant is available.`);
+          log.warn(
+            `${result.failed} segment(s) skipped — Qdrant deletion failed. Re-run when Qdrant is available.`
+          );
         }
       }
     });
@@ -184,7 +143,7 @@ Examples:
   memory
     .command("query <text>")
     .description(
-      "Run a memory recall query and print the injected memory payload",
+      "Run a memory recall query and print the injected memory payload"
     )
     .option("-c, --conversation <id>", "Optional conversation ID")
     .addHelpText(
@@ -205,7 +164,7 @@ context-aware recall. If omitted, the most recent conversation is used.
 Examples:
   $ assistant memory query "What is the project deadline?"
   $ assistant memory query "preferred communication style" --conversation conv_abc123
-  $ assistant memory query "API rate limits"`,
+  $ assistant memory query "API rate limits"`
     )
     .action(async (text: string, opts?: { conversation?: string }) => {
       initializeDb();
@@ -215,18 +174,21 @@ Examples:
         conversationId = latest?.id ?? "";
       }
       const result = await queryMemory(text, conversationId ?? "");
-      if (result.degraded) {
-        log.info(`Memory degraded: ${result.reason ?? "unknown reason"}`);
-      }
-      log.info(`Semantic hits: ${result.semanticHits}`);
-      log.info(`Merged count: ${result.mergedCount}`);
-      log.info(`Injected tokens: ${result.injectedTokens}`);
-      log.info(`Latency: ${result.latencyMs}ms`);
-      if (result.injectedText.length > 0) {
+      log.info(`Results: ${result.results.length}`);
+      log.info(`Mode: ${result.mode}`);
+      if (result.results.length > 0) {
         log.info("");
-        log.info(result.injectedText);
+        for (const r of result.results) {
+          log.info(
+            `[${r.type}] (confidence: ${r.confidence.toFixed(
+              2
+            )}, score: ${r.score.toFixed(3)})`
+          );
+          log.info(r.content);
+          log.info("");
+        }
       } else {
-        log.info("No memory injected.");
+        log.info("No results found.");
       }
     });
 
@@ -246,7 +208,7 @@ status" to monitor job progress.
 
 Examples:
   $ assistant memory rebuild-index
-  $ assistant memory status`,
+  $ assistant memory status`
     )
     .action(() => {
       initializeDb();
@@ -257,18 +219,15 @@ Examples:
   memory
     .command("re-extract")
     .description(
-      "Re-extract memories from conversations using the latest extraction prompt",
+      "Re-extract memories from conversations using the latest extraction prompt"
     )
     .option(
       "-c, --conversation <id>",
       "Target a specific conversation by ID (repeatable)",
       (val: string, prev: string[]) => [...prev, val],
-      [] as string[],
-    )
-    .option(
-      "-t, --top <n>",
-      "Auto-select top N conversations by message count",
+      [] as string[]
     )
+    .option("-t, --top <n>", "Auto-select top N conversations by message count")
     .option("--dry-run", "Show what would be re-extracted without doing it")
     .addHelpText(
       "after",
@@ -278,7 +237,7 @@ extraction prompt. This is useful after updating the extraction prompt
 (e.g. importance scoring rework) to re-score and re-extract memories
 from historically important conversations.
 
-The command resets extraction checkpoints so the batch extraction handler
+The command resets extraction checkpoints so the graph extraction handler
 re-processes all messages. Existing memories are provided as supersession
 context — the new extraction can supersede old flat-fact memories with
 richer, properly-scored replacements.
@@ -289,14 +248,10 @@ background worker).
 Examples:
   $ assistant memory re-extract --top 20
   $ assistant memory re-extract --conversation conv_abc123
-  $ assistant memory re-extract --top 10 --dry-run`,
+  $ assistant memory re-extract --top 10 --dry-run`
     )
     .action(
-      (opts: {
-        conversation?: string[];
-        top?: string;
-        dryRun?: boolean;
-      }) => {
+      (opts: { conversation?: string[]; top?: string; dryRun?: boolean }) => {
         initializeDb();
 
         const targets = [];
@@ -333,7 +288,7 @@ Examples:
 
         if (targets.length === 0) {
           log.info(
-            "No targets specified. Use --conversation <id> or --top <n>.",
+            "No targets specified. Use --conversation <id> or --top <n>."
           );
           return;
         }
@@ -342,9 +297,7 @@ Examples:
         log.info(`\nRe-extraction targets (${targets.length}):`);
         for (const t of targets) {
           const title = t.title ?? "(untitled)";
-          log.info(
-            `  ${t.conversationId}  ${t.messageCount} msgs  "${title}"`,
-          );
+          log.info(`  ${t.conversationId}  ${t.messageCount} msgs  "${title}"`);
         }
 
         if (opts.dryRun) {
@@ -354,8 +307,8 @@ Examples:
 
         const { jobIds } = requestReextract(targets);
         log.info(
-          `\nQueued ${jobIds.length} re-extraction job(s). The assistant will process them in the background.`,
+          `\nQueued ${jobIds.length} re-extraction job(s). The assistant will process them in the background.`
         );
-      },
+      }
     );
 }

package/src/cli/commands/notifications.ts

@@ -10,6 +10,10 @@ import {
   NOTIFICATION_SOURCE_EVENT_NAMES,
 } from "../../notifications/signal.js";
 import type { NotificationChannel } from "../../notifications/types.js";
+import {
+  initAuthSigningKey,
+  resolveSigningKey,
+} from "../../runtime/auth/token-service.js";
 import { initializeDb } from "../db.js";
 import { log } from "../logger.js";
 import { shouldOutputJson, writeOutput } from "../output.js";
@@ -159,7 +163,7 @@ Examples:
           visibleInSourceNow: boolean;
           deadlineAt?: string;
           preferredChannels?: string;
-          conversationId?: string;
+          sessionId?: string;
           dedupeKey?: string;
         },
         cmd: Command,
@@ -251,8 +255,9 @@ Examples:
           }
 
           initializeDb();
+          initAuthSigningKey(resolveSigningKey());
 
-          const sourceContextId = opts.conversationId ?? `cli-${Date.now()}`;
+          const sourceContextId = opts.sessionId ?? `cli-${Date.now()}`;
 
           const result = await emitNotificationSignal({
             sourceEventName: opts.sourceEventName,

package/src/cli/commands/oauth/__tests__/connect.test.ts

@@ -88,7 +88,7 @@ mock.module("../../../../platform/client.js", () => ({
 }));
 
 mock.module("../../../../util/browser.js", () => ({
-  openInBrowser: (url: string) => {
+  openInHostBrowser: async (url: string) => {
     mockOpenInBrowserCalls.push(url);
   },
 }));
@@ -127,7 +127,7 @@ mock.module("../shared.js", () => ({
         JSON.stringify({
           ok: false,
           error:
-            "Platform prerequisites not met (not logged in or missing assistant ID)",
+            "Not connected to Vellum platform. Run `vellum platform connect` to connect first.",
         }) + "\n",
       );
       return null;

package/src/cli/commands/oauth/__tests__/disconnect.test.ts

@@ -112,7 +112,7 @@ mock.module("../../../../platform/client.js", () => ({
 }));
 
 mock.module("../../../../util/browser.js", () => ({
-  openInBrowser: () => {},
+  openInHostBrowser: async () => {},
 }));
 
 mock.module("../../../../util/logger.js", () => ({
@@ -162,7 +162,7 @@ mock.module("../shared.js", () => ({
         JSON.stringify({
           ok: false,
           error:
-            "Platform prerequisites not met (not logged in or missing assistant ID)",
+            "Not connected to Vellum platform. Run `vellum platform connect` to connect first.",
         }) + "\n",
       );
       return null;

package/src/cli/commands/oauth/__tests__/mode.test.ts

@@ -34,6 +34,10 @@ let mockSetNestedValueCalls: Array<{
 
 let mockConfigServices: Record<string, unknown> = {};
 
+let mockRequirePlatformConnection: (
+  cmd: unknown,
+) => Promise<boolean> = async () => true;
+
 // ---------------------------------------------------------------------------
 // Mocks
 // ---------------------------------------------------------------------------
@@ -120,6 +124,8 @@ mock.module("../shared.js", () => ({
   isManagedMode: () => false,
   getManagedServiceConfigKey: (key: string) =>
     mockGetManagedServiceConfigKey(key),
+  requirePlatformConnection: (cmd: unknown) =>
+    mockRequirePlatformConnection(cmd),
   requirePlatformClient: async (_cmd: Command) => {
     if (
       !mockPlatformClientResult ||
@@ -130,7 +136,7 @@ mock.module("../shared.js", () => ({
         JSON.stringify({
           ok: false,
           error:
-            "Platform prerequisites not met (not logged in or missing assistant ID)",
+            "Not connected to Vellum platform. Run `vellum platform connect` to connect first.",
         }) + "\n",
       );
       return null;
@@ -244,6 +250,7 @@ describe("assistant oauth mode", () => {
     mockSaveRawConfigCalls = [];
     mockSetNestedValueCalls = [];
     mockConfigServices = {};
+    mockRequirePlatformConnection = async () => true;
     process.exitCode = 0;
   });
 

package/src/cli/commands/oauth/__tests__/status.test.ts

@@ -71,7 +71,7 @@ mock.module("../../../../platform/client.js", () => ({
 }));
 
 mock.module("../../../../util/browser.js", () => ({
-  openInBrowser: () => {},
+  openInHostBrowser: async () => {},
 }));
 
 mock.module("../../../../util/logger.js", () => ({
@@ -107,7 +107,7 @@ mock.module("../shared.js", () => ({
         JSON.stringify({
           ok: false,
           error:
-            "Platform prerequisites not met (not logged in or missing assistant ID)",
+            "Not connected to Vellum platform. Run `vellum platform connect` to connect first.",
         }) + "\n",
       );
       return null;

package/src/cli/commands/oauth/connect.ts

@@ -2,6 +2,7 @@ import { createServer, type Server } from "node:http";
 
 import type { Command } from "commander";
 
+import { getIsContainerized } from "../../../config/env-registry.js";
 import { orchestrateOAuthConnect } from "../../../oauth/connect-orchestrator.js";
 import {
   getAppByProviderAndClientId,
@@ -9,7 +10,7 @@ import {
   getProvider,
 } from "../../../oauth/oauth-store.js";
 import { renderOAuthCompletionPage } from "../../../security/oauth-completion-page.js";
-import { openInBrowser } from "../../../util/browser.js";
+import { openInHostBrowser } from "../../../util/browser.js";
 import { getSecureKeyViaDaemon } from "../../lib/daemon-credential-client.js";
 import { getCliLogger } from "../../logger.js";
 import { shouldOutputJson, writeOutput } from "../../output.js";
@@ -178,15 +179,23 @@ Examples:
 
             // When opening the browser, start a local server to show a nice
             // completion page instead of redirecting to the platform website.
+            //
+            // In containerized mode the loopback server is unreachable from
+            // the host browser, so redirect to the platform's own completion
+            // page instead.
             let redirectServer:
               | { redirectUrl: string; cleanup: () => void }
               | undefined;
             if (opts.browser !== false) {
-              try {
-                redirectServer = await startManagedRedirectServer(provider);
-                body.redirect_after_connect = redirectServer.redirectUrl;
-              } catch {
-                // Non-fatal — fall back to platform default redirect
+              if (getIsContainerized()) {
+                body.redirect_after_connect = "/account/oauth/desktop-complete";
+              } else {
+                try {
+                  redirectServer = await startManagedRedirectServer(provider);
+                  body.redirect_after_connect = redirectServer.redirectUrl;
+                } catch {
+                  // Non-fatal — fall back to platform default redirect
+                }
               }
             }
 
@@ -199,9 +208,14 @@ Examples:
 
               if (!response.ok) {
                 const errorText = await response.text().catch(() => "");
-                writeError(
-                  `Platform returned HTTP ${response.status}${errorText ? `: ${errorText}` : ""}`,
-                );
+                const baseMsg = `Platform returned HTTP ${response.status}${errorText ? `: ${errorText}` : ""}`;
+                if (response.status === 401 || response.status === 403) {
+                  writeError(
+                    `${baseMsg}. Your platform session may have expired. Run \`vellum platform connect\` to reconnect.`,
+                  );
+                } else {
+                  writeError(baseMsg);
+                }
                 return;
               }
 
@@ -229,7 +243,7 @@ Examples:
                 }
                 const snapshotIds = new Set(snapshotEntries.map((e) => e.id));
 
-                openInBrowser(result.connect_url);
+                await openInHostBrowser(result.connect_url);
 
                 if (!jsonMode) {
                   log.info(
@@ -374,7 +388,7 @@ Examples:
               clientSecret,
               callbackTransport: opts.callbackTransport,
               isInteractive: opts.browser !== false,
-              openUrl: opts.browser !== false ? openInBrowser : undefined,
+              openUrl: opts.browser !== false ? openInHostBrowser : undefined,
               ...(opts.scopes ? { requestedScopes: opts.scopes } : {}),
             });
 

package/src/cli/commands/oauth/mode.ts

@@ -17,6 +17,7 @@ import { shouldOutputJson, writeOutput } from "../../output.js";
 import {
   fetchActiveConnections,
   getManagedServiceConfigKey,
+  requirePlatformConnection,
 } from "./shared.js";
 
 /**
@@ -180,6 +181,12 @@ Examples:
           return;
         }
 
+        // Require platform connection when switching to managed mode
+        if (newMode === "managed") {
+          const connected = await requirePlatformConnection(cmd);
+          if (!connected) return;
+        }
+
         // Read current mode
         const services: Services = getConfig().services;
         const currentMode = services[managedKey as keyof Services].mode;

package/src/cli/commands/oauth/shared.ts

@@ -59,11 +59,20 @@ export async function requirePlatformClient(
   cmd: Command,
 ): Promise<VellumPlatformClient | null> {
   const client = await VellumPlatformClient.create();
-  if (!client || !client.platformAssistantId) {
+  if (!client) {
     writeOutput(cmd, {
       ok: false,
       error:
-        "Platform prerequisites not met (not logged in or missing assistant ID)",
+        "Not connected to Vellum platform. Run `vellum platform connect` to connect first.",
+    });
+    process.exitCode = 1;
+    return null;
+  }
+  if (!client.platformAssistantId) {
+    writeOutput(cmd, {
+      ok: false,
+      error:
+        "Connected to Vellum platform but no assistant ID is configured. Ensure the assistant is registered on the platform.",
     });
     process.exitCode = 1;
     return null;
@@ -71,6 +80,29 @@ export async function requirePlatformClient(
   return client;
 }
 
+/**
+ * Verify that the user has connected to the Vellum platform (has stored
+ * credentials). Unlike `requirePlatformClient`, this does NOT require a
+ * platform assistant ID — it only checks that credentials exist.
+ *
+ * Writes an error and sets exitCode=1 when the user is not connected.
+ */
+export async function requirePlatformConnection(
+  cmd: Command,
+): Promise<boolean> {
+  const client = await VellumPlatformClient.create();
+  if (!client) {
+    writeOutput(cmd, {
+      ok: false,
+      error:
+        "Not connected to Vellum platform. Run `vellum platform connect` to connect first.",
+    });
+    process.exitCode = 1;
+    return false;
+  }
+  return true;
+}
+
 /**
  * Fetch active platform connections for a provider. Returns the parsed entries
  * or writes an error and returns null.
@@ -94,9 +126,13 @@ export async function fetchActiveConnections(
 
   if (!response.ok) {
     if (!options?.silent) {
+      const hint =
+        response.status === 401 || response.status === 403
+          ? `. Your platform session may have expired. Run \`vellum platform connect\` to reconnect.`
+          : "";
       writeOutput(cmd, {
         ok: false,
-        error: `Platform returned HTTP ${response.status}`,
+        error: `Platform returned HTTP ${response.status}${hint}`,
       });
       process.exitCode = 1;
     }

package/src/cli/commands/platform/__tests__/connect.test.ts

@@ -30,7 +30,7 @@ mock.module("../../../lib/daemon-credential-client.js", () => ({
 
 mock.module("../../../../inbound/platform-callback-registration.js", () => ({
   resolvePlatformCallbackRegistrationContext: async () => ({
-    containerized: false,
+    isPlatform: false,
     platformBaseUrl: "",
     assistantId: "",
     hasInternalApiKey: false,

package/src/cli/commands/platform/__tests__/disconnect.test.ts

@@ -36,7 +36,7 @@ mock.module("../../../../config/env-registry.js", () => ({
 
 mock.module("../../../../inbound/platform-callback-registration.js", () => ({
   resolvePlatformCallbackRegistrationContext: async () => ({
-    containerized: false,
+    isPlatform: false,
     platformBaseUrl: "",
     assistantId: "",
     hasInternalApiKey: false,

package/src/cli/commands/platform/__tests__/status.test.ts

@@ -11,7 +11,7 @@ let mockGetSecureKeyViaDaemon: (
 let mockResolvePlatformCallbackRegistrationContext: () => Promise<
   Record<string, unknown>
 > = async () => ({
-  containerized: false,
+  isPlatform: false,
   platformBaseUrl: "",
   assistantId: "",
   hasInternalApiKey: false,
@@ -138,7 +138,7 @@ describe("assistant platform status", () => {
   beforeEach(() => {
     mockGetSecureKeyViaDaemon = async () => undefined;
     mockResolvePlatformCallbackRegistrationContext = async () => ({
-      containerized: false,
+      isPlatform: false,
       platformBaseUrl: "",
       assistantId: "",
       hasInternalApiKey: false,
@@ -158,7 +158,7 @@ describe("assistant platform status", () => {
 
     // GIVEN a containerized environment with platform configuration
     mockResolvePlatformCallbackRegistrationContext = async () => ({
-      containerized: true,
+      isPlatform: true,
       platformBaseUrl: "https://platform.vellum.ai",
       assistantId: "asst-abc-123",
       hasInternalApiKey: true,
@@ -191,7 +191,7 @@ describe("assistant platform status", () => {
 
     // AND the output contains the expected status fields
     const parsed = JSON.parse(stdout);
-    expect(parsed.containerized).toBe(true);
+    expect(parsed.isPlatform).toBe(true);
     expect(parsed.baseUrl).toBe("https://platform.vellum.ai");
     expect(parsed.assistantId).toBe("asst-abc-123");
     expect(parsed.hasInternalApiKey).toBe(true);
@@ -211,7 +211,7 @@ describe("assistant platform status", () => {
 
     // GIVEN a disconnected environment with no stored credentials
     mockResolvePlatformCallbackRegistrationContext = async () => ({
-      containerized: false,
+      isPlatform: false,
       platformBaseUrl: "",
       assistantId: "",
       hasInternalApiKey: false,