pi-llama-cpp 0.2.0 → 0.2.1

package/README.md

@@ -6,20 +6,21 @@ A [Pi Coding Agent](https://pi.dev/) extension that integrates with a running [l
 
 - **Auto-detect models** — discovers all models available on your running llama.cpp server
 - **Live status indicators** — see which models are loaded, loading, failed, sleeping, or unloaded with color-coded icons
-
-  | Icon | Status | Description |
-  |------|--------|-------------|
-  | 🟢 | Loaded | Model is active and ready to use |
-  | 🟡 | Loading | Model is currently being loaded |
-  | 🔴 | Failed | Model failed to load |
-  | 🔵 | Sleeping | Model is loaded but inactive (router mode) |
-  | ⚪ | Unloaded | Model is not loaded on the server |
-
 - **Load / unload / switch** — manage models directly from the Pi command palette
 - **Multi-model router support** — works with both single-model and multi-model llama.cpp server configurations
-- **Image model support** — detects multimodal models automatically
+- **Image capabilities detection** — detects multimodal models automatically
 - **Flexible URL resolution** — configures the server URL via project config, environment variable, or global settings
 
+### Status Indicators
+
+| Icon | Status | Description |
+|------|--------|-------------|
+| 🟢 | Loaded | Model is active and ready to use |
+| 🟡 | Loading | Model is currently being loaded |
+| 🔴 | Failed | Model failed to load |
+| 🔵 | Sleeping | Model is loaded but inactive (router mode) |
+| ⚪ | Unloaded | Model is not loaded on the server |
+
 ## Installation
 
 This package is a Pi extension. Install it with
@@ -60,9 +61,9 @@ The extension resolves the llama.cpp server URL using the following priority ord
 
 ### API Key
 
-If your llama.cpp server requires authentication, use `/login` in Pi, select the "API key" option, and choose the `Llama.cpp` provider.
+If your llama.cpp server requires authentication, use `/login` in Pi, select the "API key" option, and choose the `Llama.cpp` provider from the list.
 
-Alternatively, configure the API key in `~/.pi/agent/auth.json`:
+Alternatively, configure the API key in `~/.pi/agent/auth.json` using the provider ID `llama-server`:
 
 ```json
 {
@@ -77,15 +78,17 @@ Alternatively, configure the API key in `~/.pi/agent/auth.json`:
 
 ### Prerequisites
 
-Make sure your llama.cpp server is running with the appropriate flags. For multi-model support (model router), start the server with:
+Make sure your llama.cpp server is running with the appropriate flags.
+
+- For multi-model support (model router), start the server with:
 
 ```bash
-llama-server --models-preset path/to/presets.ini
+llama-server --models-preset path/to/presets.ini ...
 ```
 
-(You can use both `--fit-ctx` and `--ctx-size` in the preset — the extension checks both.)
+The extension reads the context size from the preset file using the `ctx-size` and/or `fit-ctx` keys.
 
-For single-model mode, a standard invocation works:
+- For single-model mode, start the server with:
 
 ```bash
 llama-server --model path/to/model.gguf --ctx-size 128000 ...
@@ -97,6 +100,8 @@ llama-server --model path/to/model.gguf --ctx-size 128000 ...
 | --------- | ------------------------------------------------------------------------------------------ |
 | `/models` | Browse your models with live status. Select a model to load, switch, or unload it.         |
 
+> **Note:** When the llama.cpp server is unreachable, `/models` is still available but displays an error notification with the configured server URL.
+
 ### Model Actions
 
 When browsing models via the `/models` command, you can:
@@ -108,6 +113,8 @@ When browsing models via the `/models` command, you can:
 - **Info** — View model details (ID, capabilities, context size)
 - **Cancel** — Cancel the current operation
 
+> **Note:** In single-model mode, only **Info** and **Cancel** are available, since there is only one model loaded on the server.
+
 ### Model Selection Event
 
 When Pi switches models (via `model_select`), the extension automatically loads the selected model on the llama.cpp server. This keeps the server in sync with the active model in Pi.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-llama-cpp",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "description": "Pi extension for llama.cpp integration. Supports both router and single modes",
   "keywords": [
     "pi",

package/src/handlers.ts

@@ -9,25 +9,30 @@ import { Status } from "./enums/status";
 import { BaseModel } from "./models/baseModel";
 
 /**
- * Defines a handler when llama-server is running
+ * Select a model from the list. Returns null if user cancels.
+ *
  * @param ctx Pi context
- * @returns The action and model, if detected
+ * @param models A list of models
+ * @returns The selected model
  */
-const modelSelectionHandler = async (
+const selectModel = async (
   ctx: ExtensionCommandContext,
   models: BaseModel[],
-): Promise<{ action: Action; model: BaseModel } | null> => {
-  // Setup the labels
+): Promise<BaseModel | null> => {
   const labels = await Promise.all(models.map((m) => m.getLabel()));
-
-  // Detect the selected model
   const choice = await ctx.ui.select(`${PROVIDER_NAME} models:`, labels);
   if (!choice) return null;
-
   const idx = labels.indexOf(choice);
-  const model = models[idx];
+  return models[idx];
+};
 
-  // Router mode actions
+/**
+ * Get available actions for a model based on its mode and status.
+ *
+ * @param model The selected model
+ * @returns
+ */
+const getActionsForModel = async (model: BaseModel): Promise<Array<Action>> => {
   const routerModeActions: Record<Status, Array<Action>> = {
     [Status.LOADED]: [Action.SWITCH, Action.UNLOAD, Action.INFO, Action.CANCEL],
     [Status.LOADING]: [Action.CANCEL],
@@ -36,7 +41,6 @@ const modelSelectionHandler = async (
     [Status.UNLOADED]: [Action.LOAD, Action.CANCEL],
   };
 
-  // Single mode actions (more limited)
   const singleModeActions: Record<Status, Array<Action>> = {
     [Status.LOADED]: [Action.INFO, Action.CANCEL],
     [Status.LOADING]: [Action.CANCEL],
@@ -45,21 +49,69 @@ const modelSelectionHandler = async (
     [Status.UNLOADED]: [Action.CANCEL],
   };
 
-  // Define the actions that the user can do
   const allActions =
     model.mode === Mode.ROUTER ? routerModeActions : singleModeActions;
 
   const status = await model.getStatus();
-  const actions = allActions[status];
+  return allActions[status];
+};
+
+/**
+ * Selects an action for a model.
+ *
+ * @param ctx Pi context
+ * @param model The selected model
+ * @param actions Possible actions to execute
+ * @returns The action, or null if user cancels
+ */
+const selectAction = async (
+  ctx: ExtensionCommandContext,
+  model: BaseModel,
+  actions: Array<Action>,
+): Promise<Action | null> => {
+  const labels = actions.map((a) => String(a));
+  const choice = await ctx.ui.select(`${model.name}`, labels);
+  if (!choice) return null;
 
-  const action = (await ctx.ui.select(`${model.name}`, actions)) as Action;
+  const idx = labels.indexOf(choice);
+  return actions[idx];
+};
 
-  // Send the selected action with the corresponding model
-  return { action, model };
+/**
+ * Handles the menu for model selection
+ * Loops: select model → select action → handle action.
+ *
+ * Escape on actions menu goes back to model selection.
+ * Escape on model selection exits.
+ *
+ * @param ctx Pi context
+ * @returns The action and model, if detected
+ */
+const modelSelectionHandler = async (
+  ctx: ExtensionCommandContext,
+  models: BaseModel[],
+): Promise<{ action: Action; model: BaseModel } | null> => {
+  while (true) {
+    // Select the model
+    const model = await selectModel(ctx, models);
+    if (!model) return null;
+
+    // Select the action
+    const actions = await getActionsForModel(model);
+    const action = await selectAction(ctx, model, actions);
+    if (action === null) {
+      // Escape key pressed => back to model selection
+      continue;
+    }
+
+    // Return the selected action and model
+    return { action, model };
+  }
 };
 
 /**
  * Handles the /models command
+ *
  * @param ctx The context used by Pi
  * @param pi The Pi extension
  */