@agent-native/core 0.147.0 → 0.148.1

package/corpus/templates/tasks/.agents/skills/frontend-design/SKILL.md

@@ -2,16 +2,19 @@
 name: frontend-design
 description: >-
   Sets the visual direction for a new or redesigned surface, with production
-  quality that avoids generic AI aesthetics. Use when building a new page,
-  app, or marketing surface, defining visual identity, or doing a design pass
-  ("make this look good"). Do not load it for routine UI edits: adding a field
-  to an existing form, fixing spacing, wiring a button, or changing copy.
+  quality that avoids generic AI aesthetics. Use when building, redesigning, or
+  cleaning up any user-facing UI, including screenshot-driven feedback, copy or
+  density reduction, settings, control placement, or a design pass ("make this
+  look good"). Do not load it only for purely mechanical wiring or formatting.
 scope: dev
 license: Complete terms in LICENSE.txt
 source: https://github.com/anthropics/skills/blob/main/skills/frontend-design/SKILL.md
 local-changes: >-
-  Description narrowed and Verification rescoped deliberately; an upstream sync
-  must not restore the broad auto-load triggers or screenshot-everything step.
+  Description deliberately widened on 2026-08-09 because the previous narrow
+  trigger excluded routine UI edits, including screenshot-driven cleanup, copy
+  and density changes, settings, and control placement. An upstream sync must
+  not re-narrow it. Verification remains deliberately rescoped; an upstream
+  sync must not restore screenshot-everything or broad browser-automation steps.
 metadata:
   internal: true
 ---
@@ -26,10 +29,12 @@ The user may ask for a component, page, full app, dashboard, marketing surface,
 
 Before coding, decide:
 
-- **Purpose**: What workflow does this surface make easier? What is the primary action?
+- **Purpose**: What workflow does this surface make easier? What does the user
+  need to understand, decide, or do next?
 - **Audience**: Who will use it repeatedly, and what should feel fast, calm, playful, premium, editorial, technical, or utilitarian?
 - **Tone**: Choose a concrete aesthetic direction: refined minimal, dense operations console, editorial, playful, industrial, warm handmade, high-contrast data tool, etc.
-- **Information hierarchy**: What must be visible in the first five seconds, and what should be progressively disclosed?
+- **Information hierarchy**: What must be visible to orient and act, and what
+  can wait until context or intent makes it relevant?
 - **Differentiation**: What makes this feel designed for this exact domain?
 
 Then implement working code that is cohesive, accessible, responsive, and polished in small details: typography, spacing, copy, motion, empty states, loading states, focus states, and error states.
@@ -50,46 +55,80 @@ before selecting its accent family. Shared behavior and semantic token names
 should stay consistent; palette, density, composition, type contrast, and
 shape language should not be identical by default.
 
-## Minimalism And Progressive Disclosure
-
-Default to Apple/Linear-level restraint: make the primary workflow obvious, then remove everything that does not help that workflow right now. A polished UI often has fewer visible controls, fewer borders, fewer labels, and fewer explanatory surfaces than the first reasonable implementation.
-
-### Progressive Disclosure Is A Design Requirement
-
-Treat an all-at-once interface as a defect to fix during design, not as a
-styling preference. Before coding, inventory every piece of content and action
-on the surface, then assign each one to the smallest useful visibility level:
-
-1. **Immediate** — the page title, current state, one primary action, and the
-   compact context needed to choose what to do next.
-2. **Expanded** — the details needed for the selected item or active workflow.
-3. **On demand** — advanced settings, diagnostics, credentials, metadata,
-   destructive actions, documentation, and rarely used tools.
-
-Use single-select accordions or simple disclosure rows for sibling panels when
-the user is choosing one item at a time. Inside an expanded panel, keep another
-layer for independent concerns instead of dumping every form, explanation, and
-secondary action into the first reveal. Prefer one flat panel with alignment,
-dividers, and whitespace over nested cards; provider or product icons should
-not receive decorative borders or containers unless the container communicates
-state or interaction. A collapsed row should still show the item name, status,
-and a concise summary so the user can scan the whole surface without opening
-everything.
-
-Before shipping, ask: “What can disappear until the user asks for it?” Then
-verify collapsed, expanded, loading, empty, error, and narrow-width states.
-If the first viewport contains multiple forms, repeated explanatory copy,
-documentation links, and controls for unrelated tasks, the surface has not
-passed this requirement yet.
-
-- **Start by subtracting**: Before adding a visible control, banner, toolbar row, card, or explanatory block, ask what can be removed, merged, renamed, or moved into an existing affordance.
-- **One primary action**: Each surface should have one dominant next action. Secondary actions belong in menus, popovers, command palettes, disclosure rows, or contextual hover/focus states unless they are used constantly.
-- **Progressively disclose rare work**: Advanced options, diagnostics, metadata, settings, import/export, destructive actions, and inspection tools should stay tucked away until requested. Prefer small icon triggers with tooltips, popovers, drawers, or detail panels over permanent chrome.
-- **Keep chrome quiet**: Avoid new always-visible bars, badges, callouts, helper text, and counters unless they prevent mistakes or are central to repeated use. Status can often be a dot, ring, muted count, or tooltip.
-- **Favor content over containers**: Do not wrap every section in a card. Use whitespace, alignment, typography, dividers, and full-width bands before adding boxes.
-- **Design for repeated use**: Production app UI should feel calm after the hundredth use. If a control shouts, animates, explains itself, or occupies a full row for an occasional action, hide or compress it.
-- **Make absence intentional**: Empty states should be sparse and action-oriented. Do not fill blank space with marketing copy, decorative art, or lists of features just because the screen feels empty.
-- **Use familiar primitives**: Icon buttons need clear tooltips. Menus, popovers, tabs, switches, and segmented controls should carry complexity instead of exposing every option at once.
+## High-Signal Minimalism And Progressive Disclosure
+
+Minimalism means fewer competing kinds of UI, not less useful information. Keep
+surfaces dense and scannable: reduce redundant prose, decorative containers,
+borders, badges, accent colors, and persistent controls before removing content
+that helps the user orient, compare, decide, or recover. The reference class is
+calm productivity software such as ChatGPT/Codex, Linear, and Vercel, not empty
+whitespace.
+
+### Product UX Philosophy
+
+Apply these principles as binding decision rules:
+
+- **Purpose over presence**: Every visible element must support orientation,
+  state comprehension, a decision, an action, or recovery. Before adding one,
+  name which; if it supports none or duplicates another element, remove or
+  defer it.
+- **Clarity over cleverness**: Use plain language, familiar patterns, explicit
+  hierarchy, and visible state. Do not make decoration or invented terminology
+  carry meaning. Preserve the labels, hit targets, focus states, and recovery
+  paths required for accessible use.
+- **Simple first, powerful on demand**: Keep the current task and state in the
+  default presentation. Put advanced, rare, diagnostic, and secondary work
+  behind contextual disclosure, menus, or Settings. Do not turn a discoverability
+  concern into permanent chrome, and do not hide information needed for the
+  current decision.
+- **Hierarchy over chrome**: Use typography, spacing, alignment, and restrained
+  emphasis to organize information. Treat prose, controls, borders, cards,
+  badges, and accent colors as a competition budget: reduce competing kinds of
+  elements before reducing useful information. When a surface has one clear
+  next decision, give it dominant emphasis and move secondary actions out of
+  the default path. Do not wrap every section in a card or add borders for
+  grouping alone; use spacing, alignment, typography, or dividers first. A
+  container must communicate grouping, state, or interaction.
+- **Coherence over novelty**: Reuse the product's primitives, interaction
+  patterns, and language. When given a screenshot or reference product, match
+  its underlying hierarchy, density, and behavior; do not copy decoration or
+  introduce a second visual language without a product reason.
+- **Quiet continuity**: Feedback must be immediate, contextual, and honest.
+  Preserve the user's place and layout, make loading, empty, and error states
+  understandable, and use motion only to explain a change.
+- **Respect repeated use**: Optimize for the hundredth use, not the first
+  impression. If a control repeats an explanation, occupies persistent space
+  for an occasional task, or turns routine work into ceremony, remove it or
+  move it to the contextual surface that owns it.
+
+### Progressive Disclosure And Information Timing
+
+Expose information and controls according to relevance, frequency, and
+consequence. Before coding, inventory the content and actions and assign them to
+the smallest useful visibility layer:
+
+1. **Orient and act**: location, current state, and context needed for the next
+   meaningful choice.
+2. **Support active work**: details and controls for the selected item or
+   current workflow.
+3. **Available on demand**: advanced settings, diagnostics, credentials,
+   metadata, destructive actions, documentation, and rare tools.
+
+Use a disclosure pattern that matches the task. Keep independent concerns easy
+to scan inside an expanded area instead of dumping every form, explanation,
+and secondary action into the first reveal. A collapsed row must retain its
+label and essential state; remove explanatory subtext unless it changes the
+next decision. For configuration with a known current state, show that state
+first and expose editing contextually instead of defaulting to a full form.
+
+Before shipping, verify collapsed, expanded, loading, empty, error, and
+narrow-width states. The default state fails review when its first viewport
+contains multiple unrelated forms, repeated explanatory copy, documentation
+links, or controls for unrelated tasks. It also fails when useful information
+has been replaced with empty whitespace or extra clicks solely to look minimal.
+The default state passes when users can identify their location, current state,
+current task, and relevant next choice without reading documentation, while
+the useful information remains dense and scannable.
 
 ## Aesthetic Guidelines
 
@@ -166,26 +205,28 @@ passed this requirement yet.
 
 ### Product Surface Review
 
-Before shipping a new app or a substantial redesign, review the first viewport
-as an operator would use it repeatedly:
-
-- Show one clear job title, the current state, and one primary next action.
-- Put optional inputs, provider choices, diagnostics, long explanations, and
-  secondary actions behind a disclosure, a later step, a menu, or the agent
-  sidebar. Do not put multiple unrelated forms side by side by default.
-- Prefer a short step flow or a focused page per domain job over one giant
-  dashboard. A route can have several states, but it should reveal one state
-  at a time and preserve the user's progress.
+Before shipping a new app or a substantial redesign, review the surface as an
+operator would use it repeatedly:
+
+- Can users identify where they are, what state they are in, and what
+  meaningful choices or next steps are available?
+- Remove competing or redundant elements, then place optional inputs, provider
+  choices, diagnostics, long explanations, and secondary actions where they
+  remain discoverable without crowding the current task.
+- Match the composition to the workflow. Use a focused page or step flow when
+  the work is sequential; use an overview when comparing or monitoring several
+  things is genuinely the job. Preserve the user's progress in either case.
 - Remove generic hero copy, feature tours, repeated helper text, nested cards,
   status-chip soup, and decorative AI treatment before adding more styling.
-- For review flows, stack the source/original first and the generated or safe
-  result second by default. Use side-by-side comparison only when the content
-  is short enough to scan without excessive horizontal reading.
+- For review flows, choose stacking, side-by-side comparison, or another
+  composition based on content length, scanability, and the user's comparison
+  task.
 - Compare the result with sibling apps. Shared toolkit behavior should feel
   consistent, but a repeated palette, hero composition, type pairing, and
   radius language without a product reason is visual drift, not consistency.
-- Check the result at the target desktop width and a narrow width. If the first
-  viewport feels like documentation instead of a tool, subtract again.
+- Check the result with realistic content at the target desktop width and a
+  narrow width. If it feels like documentation instead of a tool, reduce noise
+  or clarify the hierarchy.
 
 ## shadcn/ui Design Rules
 
@@ -203,18 +244,24 @@ as an operator would use it repeatedly:
   not show generic "Loading..." text for content loads; reserve `Spinner` for
   brief mutations, uploads, and progress actions. Use the app's existing
   loading primitive when it is a genuine design-system adapter. Empty states
-  should have one clear next action.
+  should communicate the state and offer the appropriate next step or small set
+  of choices.
 
 ## Anti-Patterns
 
-Avoid:
+Avoid patterns that add visual noise, obscure state, or compete with the user's
+task without a clear product reason:
 
 - Generic AI aesthetics: purple gradients, glassy cards everywhere, vague sparkle language, decorative blobs, and context-free hero sections.
 - Custom reimplementations of shadcn primitives.
 - Raw color overrides on shared components when semantic tokens or variants would work.
-- New always-visible controls for rare actions. Prefer menus, popovers, sheets, tabs, collapsibles, or advanced sections.
-- Full-width banners, persistent helper rows, decorative cards, or explanatory chrome for status that could be a compact affordance.
-- Treating progressive disclosure as optional. If a control is not part of the main daily workflow, hide it until context, hover, focus, or explicit user intent makes it relevant.
+- New always-visible controls for rare actions when a contextual surface would
+  preserve discoverability and reduce noise.
+- Full-width banners, persistent helper rows, decorative cards, or explanatory
+  chrome for status that could be communicated more clearly with less weight.
+- Hiding essential meaning or operability behind progressive disclosure. Defer
+  complexity when it is contextual or infrequent, but preserve labels, focus,
+  hit targets, accessibility, and recovery paths.
 - UI cards nested inside other cards.
 - Text or icons that resize or shift fixed-format UI on hover/loading.
 

package/corpus/templates/tasks/app/components/layout/HeaderActions.tsx

@@ -1,84 +1 @@
-import {
-  useEffect,
-  useSyncExternalStore,
-  type ReactNode,
-  type FC,
-} from "react";
-
-/**
- * External store for the page's header title + actions. We use an external
- * store (not React context) so that pages can inject ReactNode without
- * subscribing to the header state themselves — subscribing would trigger a
- * re-render on every update, which in turn creates new JSX, which updates
- * the store again, which re-renders… an infinite loop.
- *
- * Only <Header /> reads the store via useSyncExternalStore; pages only write.
- */
-
-type Listener = () => void;
-
-let currentTitle: ReactNode = null;
-let currentActions: ReactNode = null;
-const listeners = new Set<Listener>();
-
-function notify() {
-  for (const l of listeners) l();
-}
-
-function subscribe(l: Listener): () => void {
-  listeners.add(l);
-  return () => {
-    listeners.delete(l);
-  };
-}
-
-/** Consumed only by <Header /> — returns the current title. */
-export function useHeaderTitle(): ReactNode {
-  return useSyncExternalStore(
-    subscribe,
-    () => currentTitle,
-    () => currentTitle,
-  );
-}
-
-/** Consumed only by <Header /> — returns the current actions slot. */
-export function useHeaderActions(): ReactNode {
-  return useSyncExternalStore(
-    subscribe,
-    () => currentActions,
-    () => currentActions,
-  );
-}
-
-/**
- * Provider is now a no-op wrapper for backwards compatibility — the state
- * lives in the module-level store above. Kept as a component so callers of
- * <HeaderActionsProvider> don't need to change.
- */
-export const HeaderActionsProvider: FC<{ children: ReactNode }> = ({
-  children,
-}) => <>{children}</>;
-
-/** Mount a custom title into the app header. Cleans up on unmount. */
-export function useSetPageTitle(node: ReactNode) {
-  useEffect(() => {
-    currentTitle = node;
-    notify();
-    return () => {
-      currentTitle = null;
-      notify();
-    };
-  });
-}
-
-/** Mount ReactNode into the header's actions slot. Cleans up on unmount. */
-export function useSetHeaderActions(node: ReactNode) {
-  useEffect(() => {
-    currentActions = node;
-    notify();
-    return () => {
-      currentActions = null;
-      notify();
-    };
-  });
-}
+export * from "@agent-native/toolkit/app-shell";

package/corpus/templates/tasks/app/hooks/use-mobile.tsx

@@ -1,21 +1 @@
-import * as React from "react";
-
-const MOBILE_BREAKPOINT = 768;
-
-export function useIsMobile() {
-  const [isMobile, setIsMobile] = React.useState<boolean | undefined>(
-    undefined,
-  );
-
-  React.useEffect(() => {
-    const mql = window.matchMedia(`(max-width: ${MOBILE_BREAKPOINT - 1}px)`);
-    const onChange = () => {
-      setIsMobile(window.innerWidth < MOBILE_BREAKPOINT);
-    };
-    mql.addEventListener("change", onChange);
-    setIsMobile(window.innerWidth < MOBILE_BREAKPOINT);
-    return () => mql.removeEventListener("change", onChange);
-  }, []);
-
-  return !!isMobile;
-}
+export * from "@agent-native/toolkit/hooks";

package/corpus/templates/tasks/app/i18n/index.ts

@@ -1,34 +1,19 @@
-import { type AgentNativeI18nCatalog } from "@agent-native/core/client/i18n";
+import { createAgentNativeI18nCatalog } from "@agent-native/core/client/i18n";
 
 import enUS from "./en-US";
 
-export const i18nCatalog = {
-  sourceLocale: "en-US",
+export const i18nCatalog = createAgentNativeI18nCatalog({
   messages: enUS,
-  loadMessages: async (locale) => {
-    switch (locale) {
-      case "zh-CN":
-        return (await import("./zh-CN")).default;
-      case "zh-TW":
-        return (await import("./zh-TW")).default;
-      case "es-ES":
-        return (await import("./es-ES")).default;
-      case "fr-FR":
-        return (await import("./fr-FR")).default;
-      case "de-DE":
-        return (await import("./de-DE")).default;
-      case "ja-JP":
-        return (await import("./ja-JP")).default;
-      case "ko-KR":
-        return (await import("./ko-KR")).default;
-      case "pt-BR":
-        return (await import("./pt-BR")).default;
-      case "hi-IN":
-        return (await import("./hi-IN")).default;
-      case "ar-SA":
-        return (await import("./ar-SA")).default;
-      default:
-        return null;
-    }
+  localeLoaders: {
+    "zh-CN": () => import("./zh-CN"),
+    "zh-TW": () => import("./zh-TW"),
+    "es-ES": () => import("./es-ES"),
+    "fr-FR": () => import("./fr-FR"),
+    "de-DE": () => import("./de-DE"),
+    "ja-JP": () => import("./ja-JP"),
+    "ko-KR": () => import("./ko-KR"),
+    "pt-BR": () => import("./pt-BR"),
+    "hi-IN": () => import("./hi-IN"),
+    "ar-SA": () => import("./ar-SA"),
   },
-} satisfies AgentNativeI18nCatalog;
+});

package/dist/agent/engine/ai-sdk-engine.js

@@ -173,7 +173,8 @@ class AISDKEngine {
         this.defaultModel = config.model ?? PROVIDER_DEFAULT_MODELS[provider];
         this.supportedModels = PROVIDER_SUPPORTED_MODELS[provider];
         this.preserveCustomModels =
-            provider === "openai" && Boolean(config.baseUrl);
+            provider === "ollama" ||
+                (provider === "openai" && Boolean(config.baseUrl));
         this.capabilities = PROVIDER_CAPABILITIES[provider];
         this.apiKey =
             config.apiKey ??

package/dist/agent/engine/openai-compatible-endpoint.d.ts

@@ -1,2 +1,4 @@
 export declare const OPENAI_BASE_URL_ENV_VAR = "OPENAI_BASE_URL";
+export declare const OLLAMA_BASE_URL_ENV_VAR = "OLLAMA_BASE_URL";
+export declare function normalizeProviderBaseUrl(value: string): string;
 export declare function normalizeOpenAiBaseUrl(value: string): string;

package/dist/agent/engine/openai-compatible-endpoint.js

@@ -1,5 +1,6 @@
 export const OPENAI_BASE_URL_ENV_VAR = "OPENAI_BASE_URL";
-export function normalizeOpenAiBaseUrl(value) {
+export const OLLAMA_BASE_URL_ENV_VAR = "OLLAMA_BASE_URL";
+export function normalizeProviderBaseUrl(value) {
     const trimmed = value.trim();
     if (!trimmed) {
         throw new Error("Endpoint URL is required.");
@@ -20,3 +21,6 @@ export function normalizeOpenAiBaseUrl(value) {
     url.hash = "";
     return url.toString().replace(/\/+$/, "");
 }
+export function normalizeOpenAiBaseUrl(value) {
+    return normalizeProviderBaseUrl(value);
+}

package/dist/agent/engine/provider-endpoint-validation.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Validate a provider endpoint before a server-side model request can use it.
+ * `allowPrivate` is reserved for operator-owned deployment configuration; it
+ * must never be enabled for a user- or agent-supplied URL.
+ */
+export declare function validateProviderBaseUrl(value: string, options?: {
+    allowPrivate?: boolean;
+    allowLocalOllama?: boolean;
+}): Promise<string>;

package/dist/agent/engine/provider-endpoint-validation.js

@@ -0,0 +1,26 @@
+import { isBlockedExtensionUrlWithDns } from "../../extensions/url-safety.js";
+import { normalizeProviderBaseUrl } from "./openai-compatible-endpoint.js";
+function isLoopbackOllamaEndpoint(value) {
+    const hostname = new URL(value).hostname.toLowerCase();
+    return (hostname === "localhost" ||
+        hostname === "127.0.0.1" ||
+        hostname === "::1" ||
+        hostname === "[::1]");
+}
+/**
+ * Validate a provider endpoint before a server-side model request can use it.
+ * `allowPrivate` is reserved for operator-owned deployment configuration; it
+ * must never be enabled for a user- or agent-supplied URL.
+ */
+export async function validateProviderBaseUrl(value, options = {}) {
+    const normalized = normalizeProviderBaseUrl(value);
+    const allowLocalOllama = options.allowLocalOllama === true &&
+        process.env.NODE_ENV === "development" &&
+        isLoopbackOllamaEndpoint(normalized);
+    if (!options.allowPrivate &&
+        !allowLocalOllama &&
+        (await isBlockedExtensionUrlWithDns(normalized))) {
+        throw new Error("Endpoint URL resolves to a private/internal address — SSRF not allowed.");
+    }
+    return normalized;
+}

package/dist/agent/engine/provider-env-vars.d.ts

@@ -14,7 +14,7 @@ export interface ProviderEnvMeta {
     /** Input placeholder hint shown in the LLM picker's key field */
     placeholder: string;
 }
-export { OPENAI_BASE_URL_ENV_VAR } from "./openai-compatible-endpoint.js";
+export { OLLAMA_BASE_URL_ENV_VAR, OPENAI_BASE_URL_ENV_VAR, } from "./openai-compatible-endpoint.js";
 export declare const PROVIDER_ENV_META: Record<string, ProviderEnvMeta>;
 export declare const PROVIDER_TO_ENV: Record<string, string>;
 export declare const PROVIDER_ENV_VARS: readonly string[];

package/dist/agent/engine/provider-env-vars.js

@@ -7,7 +7,7 @@
  * (AssistantChat composer, settings env-var list, settings key form) pick
  * it up automatically.
  */
-export { OPENAI_BASE_URL_ENV_VAR } from "./openai-compatible-endpoint.js";
+export { OLLAMA_BASE_URL_ENV_VAR, OPENAI_BASE_URL_ENV_VAR, } from "./openai-compatible-endpoint.js";
 export const PROVIDER_ENV_META = {
     anthropic: {
         envVar: "ANTHROPIC_API_KEY",

package/dist/agent/engine/registry.d.ts

@@ -75,14 +75,16 @@ export declare function resolveDelegatedRunModel(engine: ModelResolvableEngine,
  * verbatim instead of normalized against the built-in catalog.
  *
  * `normalizeModelForEngine` honors a live engine's `preserveCustomModels`, but
- * that flag is only set on an AI SDK engine INSTANCE when the OpenAI provider
- * is pointed at an OpenAI-compatible gateway (a custom base URL — e.g. Ollama
- * Cloud or LiteLLM), whose model IDs are not in the built-in OpenAI catalog.
+ * that flag is only set on an AI SDK engine INSTANCE when the provider is
+ * Ollama, or when OpenAI is pointed at an OpenAI-compatible gateway (a custom
+ * base URL — e.g. Ollama Cloud or LiteLLM), whose model IDs are not in the
+ * built-in catalogs.
  * The static registry entry the settings actions pass to
  * `normalizeModelForEngine` cannot carry that runtime flag, so this async
- * helper reproduces the same decision — `ai-sdk:openai` AND a resolved base URL
- * — from the request's stored/deploy config. First-party OpenAI (no gateway)
- * returns false so an unknown/invalid model still normalizes to a supported one.
+ * helper reproduces the same decision from the request's stored/deploy config.
+ * Ollama always returns true because its local model inventory is user-defined;
+ * first-party OpenAI (no gateway) returns false so an unknown/invalid model
+ * still normalizes to a supported one.
  */
 export declare function resolveEnginePreservesCustomModels(entry: Pick<AgentEngineEntry, "name">): Promise<boolean>;
 /**

package/dist/agent/engine/registry.js

@@ -11,7 +11,8 @@ import { createRequire } from "node:module";
 import { assertCredentialStoreReadable, canUseDeployCredentialFallbackForRequest, getBuilderCredentialAuthFailure, getProviderCredentialAuthFailure, readDeployCredentialEnv, resolveBuilderCredentialsDetailed, resolveSecret, } from "../../server/credential-provider.js";
 import { getSetting } from "../../settings/store.js";
 import { getAgentAppModelDefaultForCurrentRequest } from "../app-model-defaults.js";
-import { normalizeOpenAiBaseUrl, OPENAI_BASE_URL_ENV_VAR, } from "./openai-compatible-endpoint.js";
+import { OLLAMA_BASE_URL_ENV_VAR, OPENAI_BASE_URL_ENV_VAR, } from "./openai-compatible-endpoint.js";
+import { validateProviderBaseUrl } from "./provider-endpoint-validation.js";
 const require = createRequire(import.meta.url);
 const _registry = new Map();
 const _packageAvailabilityCache = new Map();
@@ -234,20 +235,24 @@ export function resolveDelegatedRunModel(engine, options) {
  * verbatim instead of normalized against the built-in catalog.
  *
  * `normalizeModelForEngine` honors a live engine's `preserveCustomModels`, but
- * that flag is only set on an AI SDK engine INSTANCE when the OpenAI provider
- * is pointed at an OpenAI-compatible gateway (a custom base URL — e.g. Ollama
- * Cloud or LiteLLM), whose model IDs are not in the built-in OpenAI catalog.
+ * that flag is only set on an AI SDK engine INSTANCE when the provider is
+ * Ollama, or when OpenAI is pointed at an OpenAI-compatible gateway (a custom
+ * base URL — e.g. Ollama Cloud or LiteLLM), whose model IDs are not in the
+ * built-in catalogs.
  * The static registry entry the settings actions pass to
  * `normalizeModelForEngine` cannot carry that runtime flag, so this async
- * helper reproduces the same decision — `ai-sdk:openai` AND a resolved base URL
- * — from the request's stored/deploy config. First-party OpenAI (no gateway)
- * returns false so an unknown/invalid model still normalizes to a supported one.
+ * helper reproduces the same decision from the request's stored/deploy config.
+ * Ollama always returns true because its local model inventory is user-defined;
+ * first-party OpenAI (no gateway) returns false so an unknown/invalid model
+ * still normalizes to a supported one.
  */
 export async function resolveEnginePreservesCustomModels(entry) {
+    if (entry.name === "ai-sdk:ollama")
+        return true;
     if (entry.name !== "ai-sdk:openai")
         return false;
     try {
-        return Boolean(await resolveOpenAiBaseUrl());
+        return Boolean(await resolveProviderBaseUrl(OPENAI_BASE_URL_ENV_VAR));
     }
     catch {
         return false;
@@ -493,19 +498,22 @@ function engineCreateConfig(entry, apiKey, extra) {
         ...(extra ?? {}),
     };
 }
-async function resolveOpenAiBaseUrl() {
-    let raw = null;
-    try {
-        raw = await resolveSecret(OPENAI_BASE_URL_ENV_VAR);
-    }
-    catch {
-        raw = null;
-    }
-    if (!raw &&
-        canUseDeployCredentialFallbackForRequest(OPENAI_BASE_URL_ENV_VAR)) {
-        raw = readDeployCredentialEnv(OPENAI_BASE_URL_ENV_VAR);
-    }
-    return raw ? normalizeOpenAiBaseUrl(raw) : undefined;
+async function resolveProviderBaseUrl(envVar) {
+    const raw = await resolveSecret(envVar);
+    if (!raw && canUseDeployCredentialFallbackForRequest(envVar)) {
+        const deployValue = readDeployCredentialEnv(envVar);
+        if (!deployValue)
+            return undefined;
+        return validateProviderBaseUrl(deployValue, {
+            allowPrivate: true,
+        });
+    }
+    return raw
+        ? validateProviderBaseUrl(raw, {
+            allowLocalOllama: envVar === OLLAMA_BASE_URL_ENV_VAR &&
+                process.env.NODE_ENV === "development",
+        })
+        : undefined;
 }
 /**
  * A Builder connection we could not read is not a missing connection. Throwing
@@ -609,12 +617,17 @@ async function engineCreateConfigForEntry(entry, apiKey, extra, preferResolvedCr
                     : undefined;
         }
     }
-    if (entry.name === "ai-sdk:openai") {
+    if (entry.name === "ai-sdk:openai" || entry.name === "ai-sdk:ollama") {
         if (typeof safeExtra.baseURL === "string" && safeExtra.baseUrl == null) {
-            safeExtra.baseUrl = normalizeOpenAiBaseUrl(safeExtra.baseURL);
+            safeExtra.baseUrl = await validateProviderBaseUrl(safeExtra.baseURL, {
+                allowLocalOllama: entry.name === "ai-sdk:ollama" &&
+                    process.env.NODE_ENV === "development",
+            });
         }
         if (safeExtra.baseUrl == null) {
-            const baseUrl = await resolveOpenAiBaseUrl();
+            const baseUrl = await resolveProviderBaseUrl(entry.name === "ai-sdk:ollama"
+                ? OLLAMA_BASE_URL_ENV_VAR
+                : OPENAI_BASE_URL_ENV_VAR);
             if (baseUrl)
                 safeExtra.baseUrl = baseUrl;
         }

package/dist/agent/tool-search.d.ts

@@ -4,6 +4,7 @@ type ToolSearchArgs = {
     query?: unknown;
     limit?: unknown;
     includeSchemas?: unknown;
+    readOnlyOnly?: unknown;
 };
 type ToolParameterSummary = {
     name: string;