pi-tokenburn 0.7.4 → 0.8.0

package/README.md

@@ -17,7 +17,7 @@ Totals match the `tokenburn` CLI exactly.
 
 ## Features
 
-- **Status bar:** `🔥 1.2M ($5.10) [today] · 52k/h · 95% hit · ~$12.40 saved`: tokens, cost, burn rate, cache hit ratio and estimated cache savings for the chosen period (each extra toggleable). Turns red with `OVER BUDGET` past your limit.
+- **Status bar:** `🔥 1.2M ($5.10) [today] · 52k/h · 95% hit · ~$12.40 saved`: tokens, cost, burn rate, cache hit ratio and estimated savings (prompt cache, plus [pi-context-diet](https://github.com/sorinirimies/pi-context-diet) and [pi-squeeze](https://github.com/sorinirimies/pi-squeeze) when installed) for the chosen period (each extra toggleable). Turns red with `OVER BUDGET` past your limit.
 - **Tables:** usage per model, provider, project and session, for any period, with cache hit %, savings, cost and share.
 - **Filter and export:** limit everything to a provider, model or project; write csv or json.
 - **Pricing:** a price table fills in (or overrides) costs the logs don't have.
@@ -61,7 +61,7 @@ pi install git:github.com/sorinirimies/pi-tokenburn
 | `/tokenburn warn <pct>` | Notify once when a budget reaches `pct` percent (default 80) and once when it is exceeded |
 | `/tokenburn rate` | Toggle the burn rate in the status line (`42k/h`: tokens per hour averaged over the status period) |
 | `/tokenburn hit` | Toggle the cache hit ratio (`95% hit`) in the status line |
-| `/tokenburn saved` | Toggle the estimated cache savings (`~$12.40 saved`) in the status line |
+| `/tokenburn saved` | Toggle the estimated savings (`~$12.40 saved`: prompt cache, pi-context-diet, pi-squeeze) in the status line |
 | `/tokenburn pricing` | Where costs come from (session logs vs your price table) and which models are unpriced |
 | `/tokenburn price <model> <in> <out>` | Set a price in $ per million tokens (key `model` or `provider/model`); `price remove <model>` deletes it. `pricing mode <fill\|override>`: use the table only where the log has no cost (default), or always |
 | `/tokenburn window` | Toggle the panel below the editor |

package/extensions/command/completions.ts

@@ -41,7 +41,7 @@ const TOP_LEVEL: CompletionItem[] = [
 	{ value: "price ", label: "price", description: "Set a $/M-token price for a model" },
 	{ value: "rate", label: "rate", description: "Toggle the tokens/hour burn rate in the status line" },
 	{ value: "hit", label: "hit", description: "Toggle the cache hit ratio in the status line" },
-	{ value: "saved", label: "saved", description: "Toggle the estimated cache savings in the status line" },
+	{ value: "saved", label: "saved", description: "Toggle the estimated savings (cache, pi context-diet, pi squeeze) in the status line" },
 	{ value: "cycle", label: "cycle", description: "Rotate the status-bar period" },
 	{ value: "window", label: "window", description: "Toggle the panel below the editor" },
 	{ value: "budget ", label: "budget", description: "Set a day/week/month budget" },

package/extensions/core/aggregate.ts

@@ -1,6 +1,7 @@
 /** Aggregation of parsed turns into period totals and time series. */
+import { DIET_LOG, loadPluginLog, pluginLogPath, SQUEEZE_LOG } from "./plugin-logs.ts";
 import { loadAll, type Turn } from "./parser.ts";
-import { cacheValue, selectTurns, type UsageOptions } from "./select.ts";
+import { cacheValue, hasFilter, selectTurns, type UsageOptions } from "./select.ts";
 import { bucketLabel, bucketStart, type Unit } from "./time.ts";
 import type { PeriodStats, TokenBurnReport } from "./types.ts";
 
@@ -14,6 +15,13 @@ const emptyStats = (): PeriodStats => ({
 	sessions: 0,
 	cacheBase: 0,
 	cacheActual: 0,
+	cacheReadPriced: 0,
+	dietTokens: 0,
+	dietRequests: 0,
+	dietUsd: 0,
+	squeezeTokens: 0,
+	squeezeOutputs: 0,
+	squeezeUsd: 0,
 });
 
 export const HOUR_MS = 3_600_000;
@@ -33,6 +41,8 @@ export async function collectUsage(
 	now: Date = new Date(),
 	includeCache = true,
 	opts?: UsageOptions,
+	/** Where the pi plugins' savings logs are (default: pi's agent dir). */
+	logDir?: string,
 ): Promise<TokenBurnReport> {
 	const startOfHour = now.getTime() - HOUR_MS;
 	const startOfDay = bucketStart(now.getTime(), "day");
@@ -59,6 +69,7 @@ export async function collectUsage(
 		const cv = cacheValue(t);
 		p.cacheBase += cv.base;
 		p.cacheActual += cv.actual;
+		if (cv.base > 0) p.cacheReadPriced += t.cacheRead;
 	};
 
 	for (const turns of await loadSelected(sessionsRoot, opts)) {
@@ -81,6 +92,35 @@ export async function collectUsage(
 		if (maxTs >= startOfHour) report.hour.sessions++;
 	}
 
+	// The pi plugins' logs are not attributed to providers, models or projects: only unfiltered views count them.
+	if (!hasFilter(opts?.filter)) {
+		const windows: [PeriodStats, number][] = [
+			[report.total, -Infinity],
+			[report.year, startOfYear],
+			[report.month, startOfMonth],
+			[report.week, startOfWeek],
+			[report.today, startOfDay],
+			[report.hour, startOfHour],
+		];
+		const [diet, squeeze] = await Promise.all([loadPluginLog(pluginLogPath(DIET_LOG, logDir)), loadPluginLog(pluginLogPath(SQUEEZE_LOG, logDir))]);
+		for (const [p, start] of windows) {
+			for (const r of diet) {
+				if (r.ts < start) continue;
+				p.dietTokens += r.tokens;
+				p.dietRequests++;
+			}
+			for (const r of squeeze) {
+				if (r.ts < start) continue;
+				p.squeezeTokens += r.tokens;
+				p.squeezeOutputs++;
+			}
+			// Priced at the window's own prices: elided context would mostly have been cache reads,
+			// squeezed bash output would have been fresh input.
+			p.dietUsd = p.cacheReadPriced > 0 ? (p.dietTokens * p.cacheActual) / p.cacheReadPriced : 0;
+			p.squeezeUsd = p.cacheReadPriced > 0 ? (p.squeezeTokens * p.cacheBase) / p.cacheReadPriced : 0;
+		}
+	}
+
 	return report;
 }
 

package/extensions/core/config.ts

@@ -23,7 +23,7 @@ export interface TokenBurnConfig {
 	showRate: boolean;
 	/** Show the cache hit ratio of the status window in the status line. */
 	showHit: boolean;
-	/** Show the estimated cache savings of the status window in the status line. */
+	/** Show the estimated savings (prompt cache, pi-context-diet, pi-squeeze) of the status window in the status line. */
 	showSaved: boolean;
 	/** Notify once when a budget reaches this percentage (1 to 100); 0 turns warnings off. */
 	warnPct: number;

package/extensions/core/plugin-logs.ts

@@ -0,0 +1,57 @@
+/**
+ * The savings logs of pi plugins, in pi's agent dir: one JSON line per saving
+ * (`{"ts":"2026-01-02T09:15:00.000Z","tokens":18400,...}`). Only `ts` and `tokens` are read.
+ *
+ * - pi-context-diet: `context-diet-savings.jsonl`, one line per request it trimmed;
+ * - pi-squeeze: `squeeze-savings.jsonl`, one line per bash output it shortened.
+ *
+ * A missing or unreadable file means "no savings"; lines that do not parse are skipped.
+ */
+import { readFile, stat } from "node:fs/promises";
+import { join } from "node:path";
+import { getConfigDir } from "./config.ts";
+
+export const DIET_LOG = "context-diet-savings.jsonl";
+export const SQUEEZE_LOG = "squeeze-savings.jsonl";
+
+export interface LogRecord {
+	ts: number;
+	tokens: number;
+}
+
+/** Where a plugin's log is: `file` in pi's agent dir. */
+export const pluginLogPath = (file: string, dir: string = getConfigDir()) => join(dir, file);
+
+export function parsePluginLog(text: string): LogRecord[] {
+	const out: LogRecord[] = [];
+	for (const line of text.split("\n")) {
+		if (!line.trim()) continue;
+		try {
+			const r = JSON.parse(line) as { ts?: unknown; tokens?: unknown };
+			const ts = typeof r.ts === "string" ? Date.parse(r.ts) : NaN;
+			const tokens = r.tokens;
+			if (Number.isFinite(ts) && typeof tokens === "number" && Number.isFinite(tokens) && tokens > 0) out.push({ ts, tokens });
+		} catch {
+			/* a torn or foreign line */
+		}
+	}
+	return out;
+}
+
+// The status line refreshes after every message: re-read a log only when it changed.
+const cache = new Map<string, { mtimeMs: number; size: number; records: LogRecord[] }>();
+
+export async function loadPluginLog(path: string): Promise<LogRecord[]> {
+	try {
+		const s = await stat(path);
+		const hit = cache.get(path);
+		if (hit && hit.mtimeMs === s.mtimeMs && hit.size === s.size) return hit.records;
+		const records = parsePluginLog(await readFile(path, "utf8"));
+		cache.set(path, { mtimeMs: s.mtimeMs, size: s.size, records });
+		return records;
+	} catch {
+		return [];
+	}
+}
+
+export const clearPluginLogCache = () => cache.clear();

package/extensions/core/types.ts

@@ -20,6 +20,16 @@ export interface PeriodStats {
 	/** What cache reads would have cost as plain input vs what they cost (turns with a known input price). */
 	cacheBase: number;
 	cacheActual: number;
+	/** Cache-read tokens of the turns with a known input price (`cacheActual` / this = the cache-read price). */
+	cacheReadPriced: number;
+	/** pi-context-diet: tokens kept out of requests, the requests, and what they would have cost as cache reads. */
+	dietTokens: number;
+	dietRequests: number;
+	dietUsd: number;
+	/** pi-squeeze: tokens cut from bash output, the outputs, and what they would have cost as input. */
+	squeezeTokens: number;
+	squeezeOutputs: number;
+	squeezeUsd: number;
 }
 
 export interface TokenBurnReport {

package/extensions/features/report/report.ts

@@ -27,6 +27,12 @@ export function renderReport(input: {
 		...(models.cacheRead + models.cacheWrite > 0
 			? [`cache: ${Math.round(cacheHitRatio(models) * 100)}% of prompt tokens read from cache${cacheSavings(models) > 0 ? ` · saved ~${formatCost(cacheSavings(models))}` : ""}`]
 			: []),
+		...(report.total.dietTokens > 0
+			? [`pi context-diet: ~${formatNum(report.total.dietTokens)} tokens kept out of ${report.total.dietRequests} requests${report.total.dietUsd > 0 ? ` · saved ~${formatCost(report.total.dietUsd)}` : ""}`]
+			: []),
+		...(report.total.squeezeTokens > 0
+			? [`pi squeeze: ~${formatNum(report.total.squeezeTokens)} tokens cut from ${report.total.squeezeOutputs} bash outputs${report.total.squeezeUsd > 0 ? ` · saved ~${formatCost(report.total.squeezeUsd)}` : ""}`]
+			: []),
 		...forecastLines(report, config, now),
 	];
 	return [

package/extensions/features/status-line/commands.ts

@@ -73,6 +73,6 @@ export const statusLineCommands: HandlersFor<"window" | "status" | "cycle" | "ca
 	async saved(_cmd, { config, persist, note }) {
 		config.showSaved = !config.showSaved;
 		await persist();
-		note(`tokenburn estimated cache savings in the status line: ${onOff(config.showSaved)}`);
+		note(`tokenburn estimated savings (cache, pi context-diet, pi squeeze) in the status line: ${onOff(config.showSaved)}`);
 	},
 };

package/extensions/features/status-line/status.ts

@@ -20,8 +20,8 @@ export function windowRate(stat: PeriodStats, w: Window, firstTs: number | undef
 	return { tokensPerHour: stat.tokens / hours, costPerHour: stat.cost / hours };
 }
 
-/** Estimated dollars saved by cache reads in a window. */
-export const statSavings = (s: PeriodStats): number => Math.max(0, s.cacheBase - s.cacheActual);
+/** Estimated dollars saved in a window: cache reads, plus what pi-context-diet and pi-squeeze kept out. */
+export const statSavings = (s: PeriodStats): number => Math.max(0, s.cacheBase - s.cacheActual) + s.dietUsd + s.squeezeUsd;
 
 /** Status-line extras for the window, in order: burn rate, cache hit ratio, cache savings (each only when enabled and meaningful). */
 export function statusExtras(report: TokenBurnReport, cfg: TokenBurnConfig, now: number): string[] {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-tokenburn",
-  "version": "0.7.4",
+  "version": "0.8.0",
   "description": "Pi extension for live token usage monitoring and budget tracking (hour / day / week / month / year / all). 0 prompt tokens.",
   "keywords": ["pi-package", "pi-extension", "tokenburn", "tokens", "analytics", "budget", "stats"],
   "license": "MIT",