pi-llama-cpp 0.8.2 → 0.9.1

package/README.md

@@ -164,7 +164,7 @@ When browsing models via the `/models` command, you can:
 ### Thinking Budgets
 
 The extension supports configurable **thinking budgets** that control how many tokens the model allocates to its reasoning/thinking process.
-This is tied to Pi's thinking level selector (off, minimal, low, medium, high, xhigh).
+This is tied to Pi's thinking level selector (off, minimal, low, medium, high, xhigh, max).
 
 | Level     | Tokens | Description                  |
 | --------- | ------ | ---------------------------- |
@@ -173,7 +173,8 @@ This is tied to Pi's thinking level selector (off, minimal, low, medium, high, x
 | `low`     | 2,048  | Light reasoning              |
 | `medium`  | 8,192  | Balanced reasoning (default) |
 | `high`    | 16,384 | Extended reasoning           |
-| `xhigh`   | -1     | Unlimited reasoning          |
+| `xhigh`   | 32,768 | Deep reasoning               |
+| `max`     | -1     | Unlimited reasoning          |
 
 User-defined budgets can override the defaults by adding a `thinkingBudgets` object to `~/.pi/agent/settings.json` (global) or `.pi/settings.json` (per-project):
 
@@ -183,12 +184,13 @@ User-defined budgets can override the defaults by adding a `thinkingBudgets` obj
     "minimal": 256,
     "low": 1024,
     "medium": 2048,
-    "high": 4096
+    "high": 4096,
+    "xhigh": 8192
   }
 }
 ```
 
-Only `minimal`, `low`, `medium`, and `high` are configurable — `off` (0) and `xhigh` (-1, unlimited) are fixed.
+Only `minimal`, `low`, `medium`, `high` and `xhigh` are configurable — `off` (0) and `max` (-1, unlimited) are fixed.
 The extension automatically injects the appropriate `thinking_budget_tokens` into each request payload based on the selected level.
 
 ### Model Selection Event
@@ -219,5 +221,6 @@ Each model exposed to Pi includes the following defaults:
 
 | Peer dependency                   | Purpose             |
 | --------------------------------- | ------------------- |
+| `@earendil-works/pi-ai`           | Pi AI SDK           |
 | `@earendil-works/pi-coding-agent` | Pi Coding Agent SDK |
 | `@earendil-works/pi-tui`          | Pi TUI SDK          |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-llama-cpp",
-  "version": "0.8.2",
+  "version": "0.9.1",
   "description": "Pi extension for llama.cpp integration. Supports router, single and legacy models. Supports multiple servers.",
   "keywords": [
     "pi",
@@ -32,6 +32,7 @@
     ]
   },
   "peerDependencies": {
+    "@earendil-works/pi-ai": "*",
     "@earendil-works/pi-coding-agent": "*",
     "@earendil-works/pi-tui": "*"
   },

package/src/constants.ts

@@ -57,5 +57,6 @@ export const DEFAULT_THINKING_BUDGETS = {
   low: 2048,
   medium: 8192,
   high: 16384,
-  xhigh: -1,
+  xhigh: 32768,
+  max: -1,
 };

package/src/managers/command.ts

@@ -177,15 +177,20 @@ export class CommandManager {
         }
       };
 
+      const onFinished = async () => {
+        cleanupProgress();
+        EventManager.resetInflightModel();
+
+        // Re-scan providers to ensure accuracy of loaded models
+        await this.serverManager.update(pi);
+
+        // Force TUI refresh so Pi picks up the updated model states
+        ctx.ui.setStatus(PROVIDER_NAME, " ");
+        ctx.ui.setStatus(PROVIDER_NAME, undefined);
+      };
+
       // Load the model without blocking the UI
-      model
-        .load()
-        .then(onSuccess)
-        .catch(onFailure)
-        .finally(() => {
-          cleanupProgress();
-          EventManager.resetInflightModel();
-        });
+      model.load().then(onSuccess).catch(onFailure).finally(onFinished);
     }
   }
 

package/src/managers/events.ts

@@ -94,7 +94,7 @@ export class EventManager {
     if (level === "off")
       return { ...payload, chat_template_kwargs: { enable_thinking: false } };
 
-    if (level === "xhigh") return payload;
+    if (level === "max") return payload;
 
     return { ...payload, thinking_budget_tokens };
   }

package/src/models/baseModel.ts

@@ -180,6 +180,7 @@ export abstract class BaseModel {
         medium: "medium",
         high: "high",
         xhigh: "xhigh",
+        max: "max",
       },
       input: await this.getCapabilities(),
       contextWindow: await this.getContextSize(),

package/src/resolver.ts

@@ -1,3 +1,4 @@
+import { ApiKeyCredential, ModelThinkingLevel } from "@earendil-works/pi-ai";
 import {
   getAgentDir,
   readStoredCredential,
@@ -10,7 +11,6 @@ import {
   DEFAULT_LLAMA_SERVER_URL,
   DEFAULT_THINKING_BUDGETS,
 } from "./constants";
-import { ThinkingLevel } from "./interfaces/levels";
 
 export class ConfigResolver {
   private warnings: string[] = [];
@@ -109,8 +109,8 @@ export class ConfigResolver {
    * Resolves API key for the provider ID using Pi's stored credentials
    */
   resolveApiKey(providerId: string): string {
-    const credential = readStoredCredential(providerId);
-    return credential?.apiKey ?? API_KEY_PLACEHOLDER;
+    const credential = readStoredCredential(providerId) as ApiKeyCredential;
+    return credential?.key ?? API_KEY_PLACEHOLDER;
   }
 
   /**
@@ -128,7 +128,7 @@ export class ConfigResolver {
    *
    * @returns Selected level
    */
-  resolveThinkingLevel(): ThinkingLevel | undefined {
+  resolveThinkingLevel(): ModelThinkingLevel | undefined {
     return this.settingsManager.getDefaultThinkingLevel();
   }
 
@@ -137,7 +137,7 @@ export class ConfigResolver {
    *
    * @returns Thinking budgets
    */
-  resolveThinkingBudgets(): Record<ThinkingLevel, number> {
+  resolveThinkingBudgets(): Record<ModelThinkingLevel, number> {
     const settingsBudgets = this.settingsManager.getThinkingBudgets() ?? {};
     const availableBudgets = {
       ...DEFAULT_THINKING_BUDGETS,

package/tests/events.test.ts

@@ -55,7 +55,8 @@ describe("EventManager.onBeforeProviderRequest", () => {
       { level: "low", expected: { thinking_budget_tokens: 2048 } },
       { level: "medium", expected: { thinking_budget_tokens: 8192 } },
       { level: "high", expected: { thinking_budget_tokens: 16384 } },
-      { level: "xhigh", expected: {} },
+      { level: "xhigh", expected: { thinking_budget_tokens: 32768 } },
+      { level: "max", expected: {} },
     ])(
       'level "$level" should return $expected',
       async ({ level, expected }) => {
@@ -216,10 +217,10 @@ describe("EventManager.onBeforeProviderRequest", () => {
       expect(result).not.toHaveProperty("thinking_budget_tokens");
     });
 
-    it("should not allow overriding 'xhigh' — no budget is injected", async () => {
-      mockSettingsManager.getDefaultThinkingLevel.mockReturnValue("xhigh");
+    it("should not inject budget for 'max' — unlimited reasoning", async () => {
+      mockSettingsManager.getDefaultThinkingLevel.mockReturnValue("max");
       mockSettingsManager.getThinkingBudgets.mockReturnValue({
-        xhigh: 1,
+        max: 1,
       } as any);
 
       const server = createMockServer({

package/tests/resolver.test.ts

@@ -163,7 +163,7 @@ describe("API key resolution", () => {
   });
 
   it("should return the apiKey when present in credential", () => {
-    mockReadStoredCredential.mockReturnValue({ apiKey: "test-api-key" });
+    mockReadStoredCredential.mockReturnValue({ key: "test-api-key" });
 
     const resolver = new ConfigResolver();
     const result = resolver.resolveApiKey("llama-server=http://127.0.0.1:8080");

package/src/interfaces/levels.ts

@@ -1,7 +0,0 @@
-export type ThinkingLevel =
-  | "off"
-  | "minimal"
-  | "low"
-  | "medium"
-  | "high"
-  | "xhigh";