pi-llama-cpp 0.14.0 → 0.15.0

package/README.md

@@ -35,11 +35,12 @@ A [Pi Coding Agent](https://pi.dev/) extension that integrates with running [lla
 
 When browsing servers via `/models servers`, each server URL is prefixed with a health indicator:
 
-| Icon | Status      | Description                                   |
-| ---- | ----------- | --------------------------------------------- |
-| 🟢   | Healthy     | Server responded successfully to health check |
-| 🟡   | Timeout     | Server health check timed out                 |
-| 🔴   | Unreachable | Server could not be reached                   |
+| Icon | Status       | Description                                   |
+| ---- | ------------ | --------------------------------------------- |
+| 🟢   | Healthy      | Server responded successfully to health check |
+| 🟡   | Timeout      | Server health check timed out                 |
+| 🔴   | Unreachable  | Server could not be reached                   |
+| ⛔   | Unauthorized | Server requires an API key                    |
 
 ## Installation
 
@@ -127,6 +128,7 @@ With this config, the servers will appear in Pi as **Llama.cpp (Local Server)**
 | `sortBy`             | string  | `"asc"` | Sort order for models (see below)                             |
 | `pollingTimeout`     | number  | `60000` | Max time (ms) to wait for model loading before giving up      |
 | `serverTimeout`      | number  | `1000`  | Timeout (ms) for server health checks and SSE probes          |
+| `showServerUrls`     | boolean | `true`  | Show `[Server: <url>]` suffix in the /models model list       |
 
 > **Note:** `serverTimeout` controls individual HTTP request timeouts (health checks, SSE probe). `pollingTimeout` controls the total wait time for a model to finish loading. Increase `serverTimeout` for slow/high-latency servers, and `pollingTimeout` for large models or slow hardware.
 
@@ -137,11 +139,11 @@ Run `/models settings` to edit the scalar settings above without hand-editing JS
 #### Server list editor
 
 Run `/models servers` to add, edit or remove entries of `llamaSettings.servers`
-without hand-editing JSON. Each server URL is prefixed with a health indicator
-(🟢 healthy, 🟡 timeout, 🔴 unreachable) that reflects the result of a
-health check against the server. Each change is written immediately to the **project**
-`.pi/settings.json` if it exists, otherwise to **global**
-`~/.pi/agent/settings.json`.
+without hand-editing JSON. Each server URL is prefixed with a status indicator
+(🟢 healthy, 🟡 timeout, 🔴 unreachable, ⛔ unauthorized) that reflects the
+result of a health check and an auth probe against the server. Each change is
+written immediately to the **project** `.pi/settings.json` if it exists,
+otherwise to **global** `~/.pi/agent/settings.json`.
 
 Changes take effect immediately after closing the editor: new servers
 register their providers, removed ones leave pi's registry right away,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-llama-cpp",
-  "version": "0.14.0",
+  "version": "0.15.0",
   "description": "Pi extension for llama.cpp integration. Supports router, single and legacy models. Supports multiple servers.",
   "keywords": [
     "pi",
@@ -38,8 +38,8 @@
   },
   "type": "module",
   "devDependencies": {
-    "@types/node": "^26.6.2",
+    "@types/node": "^26.6.3",
     "prettier-plugin-organize-imports": "^4.3.0",
-    "vitest": "^5.0.1"
+    "vitest": "^5.0.2"
   }
 }

package/src/constants.ts

@@ -76,6 +76,11 @@ export const AUTOLOAD_ON_MESSAGE = false;
  */
 export const SORT_BY = "asc";
 
+/**
+ * Default for showServerUrls — show server URLs in /models by default.
+ */
+export const SHOW_SERVER_URLS = true;
+
 /**
  * Thinking budgets to send to the server, depending on user-selected level in Pi.
  */

package/src/index.ts

@@ -11,9 +11,10 @@ import { ModelSelectEvent } from "./interfaces/events";
 import { CommandManager } from "./managers/command";
 import { EventManager } from "./managers/events";
 import { ServerManager } from "./managers/server";
-import { settings } from "./managers/settings";
+import { createSettingsManager } from "./managers/settings";
 
 export default async function (pi: ExtensionAPI) {
+  const settings = createSettingsManager();
   const serverManager = new ServerManager(settings);
   const eventManager = new EventManager(serverManager, settings);
   const commandManager = new CommandManager(serverManager, settings);

package/src/interfaces/settings.ts

@@ -132,4 +132,10 @@ export interface LlamaSettings {
    * @default "asc"
    */
   sortBy?: SortBy;
+  /**
+   * Whether to show the `[Server: <url>]` suffix in the /models model list.
+   * When `false`, only model names are shown without server annotations.
+   * @default true
+   */
+  showServerUrls?: boolean;
 }

package/src/managers/command/models.ts

@@ -10,6 +10,7 @@ import { BaseModel } from "../../models/baseModel";
 import { errorMessage } from "../../utils/errors";
 import { EventManager } from "../events";
 import type { ServerManager } from "../server";
+import type { LlamaSettingsManager } from "../settings";
 
 /**
  * Interactive model selection/action flow for the `/models` command.
@@ -24,7 +25,10 @@ import type { ServerManager } from "../server";
  * server registry without coupling the menu to command routing.
  */
 export class ModelsMenu {
-  constructor(private readonly serverManager: ServerManager) {}
+  constructor(
+    private readonly serverManager: ServerManager,
+    private readonly settings: LlamaSettingsManager,
+  ) {}
 
   /**
    * Runs the interactive model selection menu.
@@ -173,6 +177,8 @@ export class ModelsMenu {
     ctx: ExtensionCommandContext,
     models: BaseModel[],
   ): Promise<BaseModel | null> {
+    const showServerUrls = await this.settings.resolveShowServerUrls();
+
     const labels = await Promise.all(
       models.map(async (model) => ({
         label: (await model.getLabel()).trim(),
@@ -191,7 +197,8 @@ export class ModelsMenu {
     const choices = labels.map(({ label, serverUrl }) => {
       const extraPadding = 2;
       const padLen = maxLength - graphemeLength(label) + extraPadding;
-      return `${label}${" ".repeat(padLen)} [Server: ${serverUrl}]`;
+      const serverSuffix = showServerUrls ? ` [Server: ${serverUrl}]` : "";
+      return `${label}${" ".repeat(padLen)}${serverSuffix}`;
     });
 
     const choice = await ctx.ui.select(`${PROVIDER_NAME} models:`, choices);

package/src/managers/command.ts

@@ -51,7 +51,7 @@ export class CommandManager {
     private readonly serverManager: ServerManager,
     private readonly settings: LlamaSettingsManager,
   ) {
-    this.modelsMenu = new ModelsMenu(serverManager);
+    this.modelsMenu = new ModelsMenu(serverManager, settings);
   }
 
   /**
@@ -108,6 +108,11 @@ export class CommandManager {
       this.notifyNotFound(ctx, url);
     }
 
+    // Notify about other warnings (e.g. unauthorized servers)
+    for (const warning of this.serverManager.getWarnings()) {
+      ctx.ui.notify(warning, "warning");
+    }
+
     if (args === "unload") {
       const models = await this.serverManager.getAllModels();
       await Promise.all(models.map((model) => model.unload()));

package/src/managers/server.ts

@@ -1,10 +1,14 @@
-import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import type {
+  ExtensionAPI,
+  ProviderModelConfig,
+} from "@earendil-works/pi-coding-agent";
 import { ApiError } from "../api/client";
 import { API_TYPE, PROVIDER_NAME } from "../constants";
 import { ServerStatus } from "../enums/serverStatus";
 import type { SortBy } from "../interfaces/sortBy";
 import { BaseModel } from "../models/baseModel";
 import { Server } from "../server";
+import { authRequiredMessage } from "../ui/strings";
 import type { LlamaSettingsManager } from "./settings";
 
 /** Model-list comparator: negative if a sorts first, positive if b does. */
@@ -45,6 +49,7 @@ export class ServerManager {
    */
   async update(pi: ExtensionAPI, timeout?: number) {
     this.failedUrls.length = 0;
+    this.warnings.length = 0;
 
     // Surface warnings from strict URL parsing (dropped invalid entries)
     this.warnings.push(...this.settings.takeWarnings());
@@ -60,9 +65,9 @@ export class ServerManager {
     }
 
     // Unregister providers that disappeared (removed or edited away);
-    // no-op for providers that were never registered
+    // `seen` tracks all kept providerIds, so we skip those still present.
     for (const old of this.servers) {
-      if (fresh.some((f) => f.providerId === old.providerId)) continue;
+      if (seen.has(old.providerId)) continue;
       pi.unregisterProvider(old.providerId);
       // Optional chain is intentional despite the non-optional type: `sse`
       // is undefined until initialize() runs (async-constructor hack — see Server)
@@ -79,34 +84,7 @@ export class ServerManager {
 
     // Initialization and registration
     for (const server of registrableServers) {
-      try {
-        await server.initialize();
-        await this.registerProvider(server, pi);
-      } catch (err) {
-        if (err instanceof ApiError && err.type === "authentication") {
-          // Register the provider with an empty model list so the user can
-          // still configure the API key via `/login` or `auth.json`. On the
-          // next scan the provider will re-initialize and discover models.
-          // Don't add to `failedUrls` — the server IS reachable, auth just
-          // isn't configured yet, so the health indicator should stay green.
-          const message = [
-            "[pi-llama-cpp]",
-            `Server at '${server.baseUrl}' requires a valid API key.`,
-            "Configure the key via `/login` or in `~/.pi/agent/auth.json`.",
-          ].join("\n");
-          this.warnings.push(message);
-          pi.registerProvider(server.providerId, {
-            name: server.providerName,
-            baseUrl: server.apiBaseUrl,
-            api: API_TYPE,
-            apiKey: server.getApiKey(),
-            models: [],
-          });
-          continue;
-        }
-        this.failedUrls.push(server.baseUrl);
-        continue;
-      }
+      await this.registerProvider(server, pi);
     }
   }
 
@@ -151,23 +129,37 @@ export class ServerManager {
   }
 
   /**
-   * Creates a Pi provider for the given server.
+   * Initializes the server and creates a Pi provider.
+   * Handles auth errors by registering with an empty model list so the user
+   * can still configure the API key via `/login` or `auth.json`.
    *
    * @param server The server
    * @param pi The Pi API
    */
   private async registerProvider(server: Server, pi: ExtensionAPI) {
-    const { apiBaseUrl, models, providerId, providerName } = server;
-    const apiKey = server.getApiKey();
-    const modelConfigs = await Promise.all(
-      models.map((m) => m.toProviderConfig()),
-    );
+    const { apiBaseUrl, apiKey, baseUrl, models, providerId, providerName } =
+      server;
+    let modelConfigs: ProviderModelConfig[] = [];
+
+    try {
+      await server.initialize();
+      modelConfigs = await Promise.all(models.map((m) => m.toProviderConfig()));
+    } catch (err) {
+      if (err instanceof ApiError && err.type === "authentication") {
+        // Don't add to `failedUrls` — the server IS reachable, auth just
+        // isn't configured yet, so the health indicator should stay green.
+        this.warnings.push(authRequiredMessage(baseUrl, providerId));
+      } else {
+        this.failedUrls.push(baseUrl);
+        return;
+      }
+    }
 
     pi.registerProvider(providerId, {
       name: providerName,
       baseUrl: apiBaseUrl,
       api: API_TYPE,
-      apiKey: apiKey,
+      apiKey,
       models: modelConfigs,
     });
   }

package/src/managers/settings.ts

@@ -13,6 +13,7 @@ import {
   REACT_TO_MODEL_SELECT,
   SERVER_TIMEOUT,
   SETTINGS_KEY,
+  SHOW_SERVER_URLS,
   SORT_BY,
   THINKING_BUDGETS,
 } from "../constants";
@@ -226,6 +227,15 @@ export class LlamaSettingsManager {
     return (await this.getLlamaSettings()).sortBy ?? SORT_BY;
   }
 
+  /**
+   * Resolves whether to show server URLs in the /models model list.
+   *
+   * @returns `true` if server URLs should be shown
+   */
+  async resolveShowServerUrls(): Promise<boolean> {
+    return (await this.getLlamaSettings()).showServerUrls ?? SHOW_SERVER_URLS;
+  }
+
   /**
    * Persists one llamaSettings field to settings and reloads the in-memory
    * settings so resolvers see the change immediately.
@@ -262,6 +272,8 @@ export class LlamaSettingsManager {
 }
 
 /**
- * Shared singleton instance used across the extension.
+ * Creates a new LlamaSettingsManager instance.
  */
-export const settings = new LlamaSettingsManager();
+export function createSettingsManager(): LlamaSettingsManager {
+  return new LlamaSettingsManager();
+}

package/src/server.ts

@@ -52,8 +52,8 @@ export class Server {
     // in ServerManager), so no lazy fallback is needed. initialize()
     // rebuilds the client to re-resolve the API key.
     this.apiClient =
-      deps.createApiClient?.(this.getApiKey()) ??
-      new ApiClient(options.baseUrl, this.getApiKey());
+      deps.createApiClient?.(this.apiKey) ??
+      new ApiClient(options.baseUrl, this.apiKey);
   }
 
   /** Base URL of this server endpoint. */
@@ -119,10 +119,8 @@ export class Server {
   /**
    * Retrieves the API key from the config resolver.
    * Tries custom ID first, then falls back to URL-based ID.
-   *
-   * @returns The API key
    */
-  getApiKey(): string {
+  get apiKey(): string {
     // Try custom ID first
     if (this.options.customId) {
       const key = this.settings.resolveApiKey(this.options.customId);
@@ -137,10 +135,9 @@ export class Server {
    * Clears the cache first so we always fetch fresh data.
    */
   async initialize() {
-    const apiKey = this.getApiKey();
     this.apiClient =
-      this.deps.createApiClient?.(apiKey) ??
-      new ApiClient(this.baseUrl, apiKey);
+      this.deps.createApiClient?.(this.apiKey) ??
+      new ApiClient(this.baseUrl, this.apiKey);
     this.sse = this.deps.createSSEManager?.(this) ?? new SSEManager(this);
     const { data } = await this.fetchModels();
     const mode = await this.detectServerMode(data);
@@ -185,7 +182,7 @@ export class Server {
    * @returns The server status
    */
   async isReady(timeout: number): Promise<ServerStatus> {
-    return checkServerHealth(this.baseUrl, timeout, this.getApiKey());
+    return checkServerHealth(this.baseUrl, timeout, this.apiKey);
   }
 
   /**

package/src/sse/manager.ts

@@ -46,7 +46,7 @@ export class SSEManager {
     try {
       const response = await fetch(this.sseEndpoint, {
         method: "GET",
-        headers: buildAuthHeaders(this.server.getApiKey()),
+        headers: buildAuthHeaders(this.server.apiKey),
         signal: AbortSignal.timeout(await this.server.getServerTimeout()),
       });
       this.sseSupported =
@@ -78,7 +78,7 @@ export class SSEManager {
     this.modelCallbacks.set(modelId, callbacks);
 
     // Create SSE client if not already created
-    this.sseClient ??= new SSEClient(this.sseEndpoint, this.server.getApiKey());
+    this.sseClient ??= new SSEClient(this.sseEndpoint, this.server.apiKey);
 
     // Subscribe a single dispatching callback to the SSE client
     if (!this.sseSubscribers.has(modelId)) {

package/src/ui/editors/editorOptions.ts

@@ -1,4 +1,7 @@
-import type { Theme } from "@earendil-works/pi-coding-agent";
+import type {
+  ExtensionUIContext,
+  Theme,
+} from "@earendil-works/pi-coding-agent";
 import type { KeybindingsManager, TUI } from "@earendil-works/pi-tui";
 import type { LlamaServer } from "../../interfaces/settings";
 
@@ -32,6 +35,10 @@ export interface SettingsListEditorOptions {
 export interface ServerSettingsListOptions extends SettingsListEditorOptions {
   /** Timeout (ms) for health checks. Defaults to 1000ms if not provided. */
   serverTimeout?: number;
+  /** Resolves the API key for a server (used for auth probing). */
+  authResolver?: (server: LlamaServer) => Promise<string>;
+  /** UI context for notifications (used to warn about auth-required servers). */
+  ui?: ExtensionUIContext;
 }
 
 /**

package/src/ui/editors/itemBuilder.ts

@@ -30,7 +30,7 @@ export abstract class ItemBuilder<TEntry, TField extends SettingField<TEntry>> {
       description: def.description,
       currentValue: def.currentValue(entry),
     };
-    return this.decorate(def, entry, base);
+    return this.decorate(def, base);
   }
 
   /** Builds the `SettingItem` array for one entry's editable fields. */
@@ -39,9 +39,5 @@ export abstract class ItemBuilder<TEntry, TField extends SettingField<TEntry>> {
   }
 
   /** Adds field-type-specific columns (e.g. submenu or fixed values). */
-  protected abstract decorate(
-    def: TField,
-    entry: TEntry,
-    base: SettingItem,
-  ): SettingItem;
+  protected abstract decorate(def: TField, base: SettingItem): SettingItem;
 }

package/src/ui/editors/listEditor.ts

@@ -66,9 +66,6 @@ export abstract class ListEditor<
     | "emptyServers"
     | "emptyOverrideEntries";
 
-  /** Total item count (entries or servers) for selection clamping. */
-  protected abstract getCurrentCount(): number;
-
   /** Returns the id for a row at the given index. */
   protected abstract getRowId(index: number): string;
 
@@ -78,6 +75,12 @@ export abstract class ListEditor<
   /** Title for the delete confirmation dialog (e.g. "Delete server"). */
   protected abstract get deleteTitle(): string;
 
+  /** Returns the number of items in the current list. Override when the
+   * list items don't directly correspond to `options.servers`. */
+  protected getCount(): number {
+    return this.options.servers.length;
+  }
+
   // -- hooks -----------------------------------------------------------------
 
   /**
@@ -111,10 +114,7 @@ export abstract class ListEditor<
 
   handleInput(data: string): void {
     // Delegate to the open modal dialog, if any
-    if (this.activeDialog) {
-      this.activeDialog.handleInput(data);
-      return;
-    }
+    if (this.delegateToDialogInput(data)) return;
 
     // Delegate to settingsList if a submenu is open
     if (this.submenuOpen && this.settingsList) {
@@ -134,15 +134,11 @@ export abstract class ListEditor<
     }
     if (kb.matches(data, "tui.select.up")) {
       this.selectedIndex =
-        this.selectedIndex === 0
-          ? this.getCurrentCount() - 1
-          : this.selectedIndex - 1;
+        this.selectedIndex === 0 ? this.getCount() - 1 : this.selectedIndex - 1;
     }
     if (kb.matches(data, "tui.select.down")) {
       this.selectedIndex =
-        this.selectedIndex === this.getCurrentCount() - 1
-          ? 0
-          : this.selectedIndex + 1;
+        this.selectedIndex === this.getCount() - 1 ? 0 : this.selectedIndex + 1;
     }
     if (data === "a") {
       this.beginAdd();
@@ -164,7 +160,7 @@ export abstract class ListEditor<
     // Empty list hint
     if (!this.settingsList) return ["Loading..."];
     const lines = this.settingsList.render(width);
-    if (this.getCurrentCount() === 0) {
+    if (this.getCount() === 0) {
       lines[lines.length - 1] = getSettingsListTheme().hint(
         HINTS[this.emptyHintKey],
       );
@@ -174,6 +170,18 @@ export abstract class ListEditor<
 
   // -- helpers ---------------------------------------------------------------
 
+  /**
+   * If a dialog is open, delegates input to it and returns true;
+   * otherwise returns false. Used to short-circuit handleInput.
+   */
+  private delegateToDialogInput(data: string): boolean {
+    if (this.activeDialog) {
+      this.activeDialog.handleInput(data);
+      return true;
+    }
+    return false;
+  }
+
   /**
    * Rebuilds the list and places the cursor on `targetIndex`.
    * Subclasses call this after add/delete/persist to refresh the UI.
@@ -181,9 +189,10 @@ export abstract class ListEditor<
   protected async rebuildList(targetIndex: number): Promise<void> {
     const list = await this.buildSettingsList();
     this.settingsList = list;
-    const count = this.getCurrentCount();
+    const count = this.getCount();
     if (count === 0) {
       this.selectedIndex = 0;
+      this.options.tui.requestRender();
       return;
     }
     this.selectedIndex = Math.min(targetIndex, count - 1);