@vellumai/assistant 0.7.0 → 0.7.2

package/src/config/feature-flag-registry.json

@@ -9,17 +9,9 @@
       "description": "Automatically trigger conversation analysis on the same cadence as memory extraction (batch threshold, idle debounce, end-of-conversation). The analysis agent has full tool access and writes back to memory and skills without user approval.",
       "defaultEnabled": false
     },
-    {
-      "id": "browser",
-      "scope": "assistant",
-      "key": "browser",
-      "label": "Browser",
-      "description": "Enable browser skill prerequisites section in the system prompt",
-      "defaultEnabled": true
-    },
     {
       "id": "user-hosted-enabled",
-      "scope": "macos",
+      "scope": "client",
       "key": "user-hosted-enabled",
       "label": "User Hosted Enabled",
       "description": "Enable user-hosted onboarding flow",
@@ -27,7 +19,7 @@
     },
     {
       "id": "local-docker-enabled",
-      "scope": "macos",
+      "scope": "client",
       "key": "local-docker-enabled",
       "label": "Local Docker Mode",
       "description": "When enabled, the Local hosting option uses Docker under the hood for sandboxed execution, hiding the separate Docker card",
@@ -41,33 +33,17 @@
       "description": "Enable the entire email integration: email CLI commands, email channel, domain registration, sequences, email readiness probes, invite adapters, and the email-setup skill",
       "defaultEnabled": false
     },
-    {
-      "id": "app-builder-multifile",
-      "scope": "assistant",
-      "key": "app-builder-multifile",
-      "label": "App Builder Multi-file",
-      "description": "Enable multi-file TSX app creation with esbuild compilation instead of single-HTML apps",
-      "defaultEnabled": true
-    },
-    {
-      "id": "mobile-pairing",
-      "scope": "macos",
-      "key": "mobile-pairing",
-      "label": "Mobile Pairing",
-      "description": "Show the Mobile (iOS) pairing card in Settings > Account",
-      "defaultEnabled": false
-    },
     {
       "id": "settings-developer-nav",
-      "scope": "assistant",
+      "scope": "client",
       "key": "settings-developer-nav",
       "label": "Settings Developer Nav",
       "description": "Control Developer nav visibility in macOS settings",
-      "defaultEnabled": true
+      "defaultEnabled": false
     },
     {
       "id": "developer-menu-items",
-      "scope": "macos",
+      "scope": "client",
       "key": "developer-menu-items",
       "label": "Developer Menu Items",
       "description": "Show Component Gallery and Replay Onboarding in the menu bar",
@@ -105,54 +81,6 @@
       "description": "Surface credential grant and audit inspection endpoints for reviewing active grants and access logs",
       "defaultEnabled": false
     },
-    {
-      "id": "ces-managed-sidecar",
-      "scope": "assistant",
-      "key": "ces-managed-sidecar",
-      "label": "CES Managed Sidecar Transport",
-      "description": "Use managed sidecar transport for CES communication when running in a containerized environment",
-      "defaultEnabled": true
-    },
-    {
-      "id": "ces-credential-backend",
-      "scope": "assistant",
-      "key": "ces-credential-backend",
-      "label": "CES Credential Backend",
-      "description": "Route credential reads and writes through the CES process instead of accessing the encrypted store directly",
-      "defaultEnabled": true
-    },
-    {
-      "id": "settings-billing",
-      "scope": "macos",
-      "key": "settings-billing",
-      "label": "Billing Settings Tab",
-      "description": "Show the Billing tab in Settings when signed in, displaying credits balance and top-up",
-      "defaultEnabled": true
-    },
-    {
-      "id": "referral-codes",
-      "scope": "macos",
-      "key": "referral-codes",
-      "label": "Referral Codes",
-      "description": "Surface the Earn Credits referral entry points (sidebar drawer row and Billing tab button) that open the referral modal",
-      "defaultEnabled": true
-    },
-    {
-      "id": "managed-sign-in",
-      "scope": "macos",
-      "key": "managed-sign-in",
-      "label": "Managed Sign In",
-      "description": "Enable managed (organization-hosted) sign-in flow in the onboarding experience",
-      "defaultEnabled": true
-    },
-    {
-      "id": "conversation-starters",
-      "scope": "assistant",
-      "key": "conversation-starters",
-      "label": "Recommended Starts",
-      "description": "Show a curated row of recommended starting actions on the empty conversation page, ordered by relevance as confident first-click options",
-      "defaultEnabled": true
-    },
     {
       "id": "deploy-to-vercel",
       "scope": "assistant",
@@ -179,7 +107,7 @@
     },
     {
       "id": "quick-input",
-      "scope": "macos",
+      "scope": "client",
       "key": "quick-input",
       "label": "Quick Input",
       "description": "Enable the Quick Input popover on right-click of the menu bar icon",
@@ -187,43 +115,19 @@
     },
     {
       "id": "expand-completed-steps",
-      "scope": "macos",
+      "scope": "client",
       "key": "expand-completed-steps",
       "label": "Expand Completed Steps",
       "description": "Auto-expand completed tool call step groups instead of showing them collapsed",
       "defaultEnabled": false
     },
-    {
-      "id": "show-thinking-blocks",
-      "scope": "macos",
-      "key": "show-thinking-blocks",
-      "label": "Show Thinking Blocks",
-      "description": "Display the assistant's thinking/reasoning inline in chat messages as collapsible blocks",
-      "defaultEnabled": true
-    },
-    {
-      "id": "inline-skill-commands",
-      "scope": "assistant",
-      "key": "inline-skill-commands",
-      "label": "Inline Skill Command Expansion",
-      "description": "Enable secure inline skill command expansion via !`command` syntax, with version-pinned approval and sandboxed execution at skill load time",
-      "defaultEnabled": true
-    },
-    {
-      "id": "channel-voice-transcription",
-      "scope": "assistant",
-      "key": "channel-voice-transcription",
-      "label": "Channel Voice Transcription",
-      "description": "Auto-transcribe voice/audio messages received from channels (Telegram, WhatsApp) before processing",
-      "defaultEnabled": true
-    },
     {
       "id": "sounds",
       "scope": "assistant",
       "key": "sounds",
       "label": "Sounds",
       "description": "Enable the Sounds tab in Settings and all app sound playback (event sounds, random ambient sounds)",
-      "defaultEnabled": true
+      "defaultEnabled": false
     },
     {
       "id": "message-tts",
@@ -239,7 +143,7 @@
       "key": "backward-releases",
       "label": "Backward Releases",
       "description": "Show older versions in the version picker, allowing rollback to previous releases",
-      "defaultEnabled": true
+      "defaultEnabled": false
     },
     {
       "id": "voice-mode",
@@ -267,7 +171,7 @@
     },
     {
       "id": "teleport",
-      "scope": "macos",
+      "scope": "client",
       "key": "teleport",
       "label": "Teleport",
       "description": "Enable teleport UI in General settings for moving assistants between hosting environments",
@@ -275,20 +179,12 @@
     },
     {
       "id": "apple-container",
-      "scope": "macos",
+      "scope": "client",
       "key": "apple-container",
       "label": "Apple Container",
       "description": "Enable assistant sandboxing via Apple Containers on macOS 26+, providing a lightweight native sandbox without third-party dependencies",
       "defaultEnabled": false
     },
-    {
-      "id": "tool-result-truncation",
-      "scope": "assistant",
-      "key": "tool-result-truncation",
-      "label": "Post-Turn Tool Result Truncation",
-      "description": "Truncate large tool results after each assistant turn, persisting full content to disk for on-demand re-reads",
-      "defaultEnabled": true
-    },
     {
       "id": "managed-gemini-embeddings-enabled",
       "scope": "assistant",
@@ -299,36 +195,20 @@
     },
     {
       "id": "fork-from-message",
-      "scope": "macos",
+      "scope": "client",
       "key": "fork-from-message",
       "label": "Fork from Message",
       "description": "Show the 'Fork from here' option in message overflow menus",
       "defaultEnabled": false
     },
-    {
-      "id": "onboarding-pre-chat",
-      "scope": "macos",
-      "key": "onboarding-pre-chat",
-      "label": "Pre-Chat Onboarding Flow",
-      "description": "Gates the 3-screen pre-chat onboarding flow (tools, tasks/tone, name exchange) shown before the first conversation",
-      "defaultEnabled": true
-    },
     {
       "id": "home-tab",
-      "scope": "macos",
+      "scope": "client",
       "key": "home-tab",
       "label": "Home Tab",
       "description": "Replace the knowledge graph top-level tab with a new Home page showing relationship progression, facts, and capability tiers",
       "defaultEnabled": false
     },
-    {
-      "id": "channel-avatar-sync",
-      "scope": "assistant",
-      "key": "channel-avatar-sync",
-      "label": "Channel Avatar Sync",
-      "description": "Automatically sync the assistant's avatar to connected channels (Slack) when the avatar changes or a new channel is connected",
-      "defaultEnabled": true
-    },
     {
       "id": "meet",
       "scope": "assistant",
@@ -337,29 +217,21 @@
       "description": "Enables the Google Meet joining bot and the meet-join skill.",
       "defaultEnabled": false
     },
-    {
-      "id": "compaction-v2",
-      "scope": "assistant",
-      "key": "compaction-v2",
-      "label": "Compaction v2",
-      "description": "Boundary-message-based compaction pipeline with forked prompt-cache-shared summarization and microcompact pre-pass. Replaces the four-tier overflow reducer and cursor-based storage.",
-      "defaultEnabled": false
-    },
     {
       "id": "scroll-debug-overlay",
-      "scope": "macos",
+      "scope": "client",
       "key": "scroll-debug-overlay",
       "label": "Scroll Debug Overlay",
       "description": "Show a live HUD in the top-right of the conversation with scroll geometry, velocity, update rate, and anchor-preserver activity. Developer diagnostic for investigating scroll jank.",
       "defaultEnabled": false
     },
     {
-      "id": "message-height-cache",
-      "scope": "macos",
-      "key": "message-height-cache",
-      "label": "Message Height Cache",
-      "description": "Swap the transcript's LazyVStack for a plain VStack so scrollContentHeight is the true sum of row heights (no LazyVStack estimator drift). Row frames are not pinned — the earlier frame-pinning approach was removed because it caused overlap when rows grew past their first measurement (streaming, expanding thinking blocks). Tradeoff: eager layout — every row measures up-front, which can stall for many seconds to minutes on very long conversations.",
-      "defaultEnabled": true
+      "id": "coding-agents-panel",
+      "scope": "client",
+      "key": "coding-agents-panel",
+      "label": "Coding Agents Panel",
+      "description": "Gate native macOS Coding Agents panel entry points and inline ACP session deep links.",
+      "defaultEnabled": false
     },
     {
       "id": "compaction-playground",
@@ -378,11 +250,27 @@
       "defaultEnabled": false
     },
     {
-      "id": "inference-profiles",
+      "id": "account-deletion",
+      "scope": "client",
+      "key": "account-deletion",
+      "label": "Account Deletion",
+      "description": "Surfaces the user-initiated account deletion flow in client settings.",
+      "defaultEnabled": false
+    },
+    {
+      "id": "app-control",
+      "scope": "assistant",
+      "key": "app-control",
+      "label": "App Control",
+      "description": "Enable the app-control skill (per-app screenshot + raw input bypassing AX tree)",
+      "defaultEnabled": false
+    },
+    {
+      "id": "analyze-conversation",
       "scope": "assistant",
-      "key": "inference-profiles",
-      "label": "Inference Profiles",
-      "description": "Show inference profile picker in chat, active profile dropdown, Manage Profiles button, and per-task overrides link in Settings",
+      "key": "analyze-conversation",
+      "label": "Analyze Conversation",
+      "description": "Show the 'Analyze' / 'Analyze conversation' option in conversation context menus and the conversation title actions dropdown.",
       "defaultEnabled": false
     }
   ]

package/src/config/llm-callsite-catalog.ts

@@ -0,0 +1,12 @@
+import { CALL_SITE_CATALOG } from "./schemas/call-site-catalog.js";
+import type { LLMCallSite } from "./schemas/llm.js";
+
+// Compatibility wrapper for existing usage display imports. Do not define
+// labels here; call-site display metadata belongs in CALL_SITE_CATALOG.
+const LLM_CALLSITE_LABELS = new Map<LLMCallSite, string>(
+  CALL_SITE_CATALOG.map(({ id, displayName }) => [id, displayName]),
+);
+
+export function getLLMCallSiteLabel(callSite: LLMCallSite | string): string {
+  return LLM_CALLSITE_LABELS.get(callSite as LLMCallSite) ?? String(callSite);
+}

package/src/config/llm-context-resolution.ts

@@ -0,0 +1,80 @@
+import { PROVIDER_CATALOG } from "../providers/model-catalog.js";
+import { resolveCallSiteConfig } from "./llm-resolver.js";
+import {
+  type ContextWindow,
+  DEFAULT_CONTEXT_WINDOW_MAX_INPUT_TOKENS,
+  type LLMCallSite,
+  type LLMConfig,
+} from "./schemas/llm.js";
+import type { ContextWindowConfig } from "./types.js";
+
+export interface EffectiveContextWindow {
+  provider: string;
+  model: string;
+  enabled: boolean;
+  maxInputTokens: number;
+  modelMaxInputTokens: number;
+  defaultInputTokens: number;
+  compactThreshold: number;
+  summaryBudgetRatio: number;
+  targetBudgetRatio: number;
+  overflowRecovery: ContextWindow["overflowRecovery"];
+  isLongContextEnabled: boolean;
+  maxOutputTokens?: number;
+}
+
+export function resolveEffectiveContextWindow({
+  llm,
+  callSite,
+  overrideProfile,
+}: {
+  llm: LLMConfig;
+  callSite: LLMCallSite;
+  overrideProfile?: string;
+}): EffectiveContextWindow {
+  const resolved = resolveCallSiteConfig(callSite, llm, { overrideProfile });
+  const catalogModel = PROVIDER_CATALOG.find(
+    (provider) => provider.id === resolved.provider,
+  )?.models.find((model) => model.id === resolved.model);
+
+  const modelMaxInputTokens =
+    catalogModel?.contextWindowTokens ??
+    DEFAULT_CONTEXT_WINDOW_MAX_INPUT_TOKENS;
+  const defaultInputTokens =
+    catalogModel?.defaultContextWindowTokens ??
+    DEFAULT_CONTEXT_WINDOW_MAX_INPUT_TOKENS;
+  const maxInputTokens = Math.min(
+    resolved.contextWindow.maxInputTokens,
+    modelMaxInputTokens,
+  );
+
+  return {
+    provider: resolved.provider,
+    model: resolved.model,
+    enabled: resolved.contextWindow.enabled,
+    maxInputTokens,
+    modelMaxInputTokens,
+    defaultInputTokens,
+    compactThreshold: resolved.contextWindow.compactThreshold,
+    summaryBudgetRatio: resolved.contextWindow.summaryBudgetRatio,
+    targetBudgetRatio: resolved.contextWindow.targetBudgetRatio,
+    overflowRecovery: resolved.contextWindow.overflowRecovery,
+    isLongContextEnabled: maxInputTokens > defaultInputTokens,
+    maxOutputTokens: catalogModel?.maxOutputTokens,
+  };
+}
+
+export function contextWindowConfigFromEffective(
+  base: ContextWindowConfig,
+  effective: EffectiveContextWindow,
+): ContextWindowConfig {
+  return {
+    ...base,
+    enabled: effective.enabled,
+    maxInputTokens: effective.maxInputTokens,
+    targetBudgetRatio: effective.targetBudgetRatio,
+    compactThreshold: effective.compactThreshold,
+    summaryBudgetRatio: effective.summaryBudgetRatio,
+    overflowRecovery: effective.overflowRecovery,
+  };
+}

package/src/config/llm-resolver.ts

@@ -3,8 +3,8 @@ import { z } from "zod";
 import {
   type LLMCallSite,
   LLMConfigBase,
-  type LLMConfigFragment,
   type LLMSchema,
+  type ProfileEntry,
 } from "./schemas/llm.js";
 
 /**
@@ -12,13 +12,24 @@ import {
  * call-site overrides, optional per-call profile, an optional ad-hoc override
  * profile, the workspace's active profile, and the required `llm.default`.
  *
- * Merge layers (low → high precedence; later layers override earlier):
+ * Merge layers (low → high precedence; later layers override earlier) for
+ * non-main-agent call sites:
  *   1. `llm.default` fields (required base)
  *   2. `llm.profiles[llm.activeProfile]` (workspace-wide active profile)
  *   3. `llm.profiles[opts.overrideProfile]` (per-call ad-hoc override)
  *   4. `llm.profiles[site.profile]` fields (call-site's named profile)
  *   5. `llm.callSites[callSite]` fields (call-site override)
  *
+ * For `mainAgent`, the selected active/conversation profile is the direct
+ * user intent for the chat loop, so profile layers intentionally sit above
+ * any static `llm.callSites.mainAgent` defaults seeded by migrations or UI
+ * settings:
+ *   1. `llm.default`
+ *   2. `llm.profiles[site.profile]`
+ *   3. `llm.callSites.mainAgent`
+ *   4. `llm.profiles[llm.activeProfile]`
+ *   5. `llm.profiles[opts.overrideProfile]`
+ *
  * Nested objects (`thinking`, `contextWindow`, and
  * `contextWindow.overflowRecovery`) are deep-merged so partial overrides at
  * any nesting level merge into — rather than replace — the corresponding
@@ -38,28 +49,51 @@ export function resolveCallSiteConfig(
 ): z.infer<typeof LLMConfigBase> {
   const layers: Mergeable[] = [llm.default as Mergeable];
 
-  // Layer: workspace-wide active profile. Silent fall-through on missing key
-  // — schema validation in LLMSchema.superRefine catches static references.
   const activeFragment =
     llm.activeProfile != null ? llm.profiles?.[llm.activeProfile] : undefined;
-  if (activeFragment != null) {
-    layers.push(activeFragment as Mergeable);
-  }
-
-  // Layer: per-call ad-hoc override (e.g. a chat conversation's pinned
-  // profile). Silent fall-through on missing key keeps the resolver pure.
   const overrideFragment =
     opts.overrideProfile != null
       ? llm.profiles?.[opts.overrideProfile]
       : undefined;
-  if (overrideFragment != null) {
-    layers.push(overrideFragment as Mergeable);
+  const site = llm.callSites?.[callSite];
+
+  if (callSite === "mainAgent") {
+    appendCallSiteLayers(layers, callSite, llm, site);
+    appendProfileLayer(layers, activeFragment);
+    appendProfileLayer(layers, overrideFragment);
+  } else {
+    appendProfileLayer(layers, activeFragment);
+    appendProfileLayer(layers, overrideFragment);
+    appendCallSiteLayers(layers, callSite, llm, site);
   }
 
-  const site = llm.callSites?.[callSite];
+  return finalize(deepMerge(...layers));
+}
+
+// ---------------------------------------------------------------------------
+// Internal helpers
+// ---------------------------------------------------------------------------
+
+type Mergeable = Record<string, unknown>;
+
+function appendProfileLayer(
+  layers: Mergeable[],
+  profile: ProfileEntry | undefined,
+): void {
+  if (profile != null) {
+    layers.push(profileConfigFragment(profile));
+  }
+}
+
+function appendCallSiteLayers(
+  layers: Mergeable[],
+  callSite: LLMCallSite,
+  llm: z.infer<typeof LLMSchema>,
+  site: z.infer<typeof LLMSchema>["callSites"][LLMCallSite] | undefined,
+): void {
   if (site != null) {
     if (site.profile != null) {
-      const profileFragment: LLMConfigFragment | undefined =
+      const profileFragment: ProfileEntry | undefined =
         llm.profiles?.[site.profile];
       if (profileFragment == null) {
         // Defensive: `LLMSchema.superRefine` already rejects unknown profile
@@ -70,22 +104,24 @@ export function resolveCallSiteConfig(
           `LLM call site "${callSite}" references undefined profile "${site.profile}"`,
         );
       }
-      layers.push(profileFragment as Mergeable);
+      layers.push(profileConfigFragment(profileFragment));
     }
     // Strip the `profile` discriminator before merging — it isn't a
     // `LLMConfigBase` field.
     const { profile: _profile, ...siteFragment } = site;
     layers.push(siteFragment as Mergeable);
   }
-
-  return finalize(deepMerge(...layers));
 }
 
-// ---------------------------------------------------------------------------
-// Internal helpers
-// ---------------------------------------------------------------------------
-
-type Mergeable = Record<string, unknown>;
+function profileConfigFragment(profile: ProfileEntry): Mergeable {
+  const {
+    source: _source,
+    label: _label,
+    description: _description,
+    ...config
+  } = profile;
+  return config as Mergeable;
+}
 
 /**
  * Returns true for objects we should recurse into during deep merge. We