pi-mtplx 0.1.6 → 0.1.8

package/README.md

@@ -36,6 +36,7 @@ Run `/mtplx` to open an interactive menu:
 | -------- | ------------- |
 | **Toggle (on/off)** | Start or stop the MTPLX server |
 | **Fan Curves** | Set the thermal profile (`default`, `smart`, `max`) |
+| **SSD Session Cache** | Enable or disable MTPLX's SSD-backed session cache for subsequent server starts (on by default) |
 | **Models** | Register or unregister models — ✓ means registered (click to unregister), ✗ means available (click to register) |
 | **Uninstall** | Remove the `mtplx` provider from Pi's config |
 

package/extensions/mtplx.ts

@@ -9,17 +9,24 @@
 import { acquire, release, stopServer } from "../src/mtplx-process.ts";
 import { getFanMode, health, setFanMode, setFanModeValue } from "../src/mtplx-client.ts";
 import { MTPLX_PROVIDER, manageModels, removeModel, removePiMtplxProvider } from "../src/model-discovery.ts";
-import { FAN_MODES, isMtplxModel, saveFanMode, type FanMode } from "../src/utils.ts";
+import { FAN_MODES, isMtplxModel, loadSsdSessionCache, saveFanMode, saveSsdSessionCache, type FanMode } from "../src/utils.ts";
 import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 
 export default function mtplxAutostart(pi: ExtensionAPI): void {
 	pi.registerCommand("mtplx", {
-		description: "MTPLX — toggle the server, pick a fan curve, manage models, or uninstall",
+		description: "MTPLX — toggle the server, configure fan and SSD cache, manage models, or uninstall",
 		handler: async (_args, ctx) => {
 			const current = await health();
 			const status = current ? "on" : "off";
 			await ctx.ui.setStatus("mtplx", `MTPLX: ${status}`);
-			const topChoices = [`Toggle (${status})`, `Fan Curves (current: ${getFanMode()})`, "Models", "Uninstall (remove provider)"];
+			const ssdSessionCache = loadSsdSessionCache();
+			const topChoices = [
+				`Toggle (${status})`,
+				`Fan Curves (current: ${getFanMode()})`,
+				`SSD Session Cache (current: ${ssdSessionCache ? "on" : "off"})`,
+				"Models",
+				"Uninstall (remove provider)",
+			];
 			const top = await ctx.ui.select("MTPLX", topChoices, undefined);
 			if (!top) return;
 			if (top.startsWith("Toggle")) {
@@ -47,6 +54,17 @@ export default function mtplxAutostart(pi: ExtensionAPI): void {
 				}
 				ctx.ui.notify(`MTPLX fan mode set to ${getFanMode()}`, "info");
 			}
+			if (top.startsWith("SSD Session Cache")) {
+				const choices = [
+					ssdSessionCache ? "Off" : "Off (current)",
+					ssdSessionCache ? "On (current)" : "On",
+				];
+				const picked = await ctx.ui.select("MTPLX — SSD session cache for future server starts", choices, undefined);
+				if (!picked) return;
+				const enabled = picked.startsWith("On");
+				saveSsdSessionCache(enabled);
+				ctx.ui.notify(`MTPLX SSD session cache will be ${enabled ? "on" : "off"} the next time the server starts.`, "info");
+			}
 			if (top.startsWith("Models")) {
 				await manageModels(ctx);
 			}
@@ -87,4 +105,4 @@ export default function mtplxAutostart(pi: ExtensionAPI): void {
 			console.error(`MTPLX cleanup on quit failed: ${error instanceof Error ? error.message : String(error)}`);
 		}
 	});
-}
+}

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "pi-mtplx",
-	"version": "0.1.6",
+	"version": "0.1.8",
 	"description": "Zero-config MTPLX integration for Pi coding agent",
 	"keywords": [
 		"pi-package",

package/src/mtplx-process.ts

@@ -21,7 +21,7 @@ import { promisify } from "node:util";
 import type { ChildProcess } from "node:child_process";
 import { getFanMode, health, setFanMode, type Health } from "./mtplx-client.ts";
 import { MTPLX_MODELS } from "./model-discovery.ts";
-import { HOST, PORT, READY_TIMEOUT_MS, POLL_MS, sleep, commandError, portIsOccupied } from "./utils.ts";
+import { HOST, PORT, READY_TIMEOUT_MS, POLL_MS, loadSsdSessionCache, sleep, commandError, portIsOccupied } from "./utils.ts";
 
 const execFileAsync = promisify(execFile);
 
@@ -110,17 +110,15 @@ export async function startServer(modelId: string): Promise<void> {
 			configured.ref,
 			"--model-id",
 			modelId,
-			"--profile",
-			"sustained",
 			"--fan-mode",
 			getFanMode(),
 			"--host",
 			HOST,
 			"--port",
 			String(PORT),
-			// Keep the KV-cache SSD cold tier off (CLI default is `on`). Do not remove.
+			// Explicitly pass the user's persisted choice; this extension defaults it to on.
 			"--ssd-session-cache",
-			"off",
+			loadSsdSessionCache() ? "on" : "off",
 		],
 		{ detached: true, stdio: "ignore" },
 	);
@@ -210,4 +208,4 @@ export function release(): void {
 		activeModel = undefined;
 		for (const resolve of idleWaiters.splice(0)) resolve();
 	}
-}
+}

package/src/utils.ts

@@ -13,6 +13,7 @@ export const POLL_MS = 500;
 
 export type FanMode = "default" | "smart" | "max";
 export const FAN_MODES: readonly FanMode[] = ["default", "smart", "max"];
+export const DEFAULT_SSD_SESSION_CACHE = true;
 
 // Fan mode ("fan curve") applied at autostart, live-updated from `/mtplx` while the
 // server runs. Persisted to disk so the choice survives Pi restarts: `fanMode` is a
@@ -41,6 +42,30 @@ export function saveFanMode(fanMode: FanMode): void {
 	}
 }
 
+// SSD session caching is deliberately opt-in for this extension. Persist the
+// choice independently of the fan setting so a user can enable it without
+// altering their existing thermal preference.
+export const SSD_SESSION_CACHE_FILE = join(homedir(), ".pi", "agent", "mtplx-ssd-session-cache.json");
+
+export function loadSsdSessionCache(): boolean {
+	try {
+		const parsed = JSON.parse(readFileSync(SSD_SESSION_CACHE_FILE, "utf8")) as { enabled?: unknown };
+		if (typeof parsed.enabled === "boolean") return parsed.enabled;
+	} catch {
+		// missing or corrupt file → fall back to the default
+	}
+	return DEFAULT_SSD_SESSION_CACHE;
+}
+
+export function saveSsdSessionCache(enabled: boolean): void {
+	try {
+		mkdirSync(join(homedir(), ".pi", "agent"), { recursive: true });
+		writeFileSync(SSD_SESSION_CACHE_FILE, JSON.stringify({ enabled }, null, 2) + "\n");
+	} catch (error) {
+		console.error(`MTPLX could not persist SSD session-cache preference: ${error instanceof Error ? error.message : String(error)}`);
+	}
+}
+
 // Derive Pi's model id from an artifact ref: "Youssofal/Qwen3.8-27B-MTPLX-Optimized-Quality"
 // → "mtplx-qwen38-27b-optimized-quality" (same scheme as the built-in entry).
 export function modelIdFromRef(ref: string): string {
@@ -98,4 +123,4 @@ export function portIsOccupied(): Promise<boolean> {
 		socket.once("timeout", () => done(false));
 		socket.once("error", () => done(false));
 	});
-}
+}