pi-antigravity-rotator 1.12.2 → 1.13.0

package/CHANGELOG.md

@@ -1,5 +1,28 @@
 # Changelog
 
+## [1.13.0] - 2026-05-19
+
+### Removed
+- **Pro Family Features**: Completely removed legacy Pro Family sharing infrastructure (Advisor recommendations, dual-window tracking, and associated UI elements) to simplify architecture for unified quota pools.
+
+### Added
+- **Quota Reset Countdown**: Added a new column to the Quota Forecast dashboard component that displays the exact time remaining until the next quota reset.
+- **Token Usage Metrics Output**: The proxy now correctly captures and forwards precise input/output token counts from the upstream API back to the client, fully enabling usage statistics reporting in compatible adapters.
+
+## [1.12.4] - 2026-05-18
+
+### Added
+- **Claude `cache_control` stripping**: Anthropic requests often include `cache_control` objects which Google Cloud Code Assist API rejects with "Extra inputs are not permitted". The proxy now safely strips `cache_control` from all system and message content blocks before forwarding them to Gemini.
+- **Claude `VALIDATED` Function Calling**: Automatically enforces `toolConfig: { functionCallingConfig: { mode: "VALIDATED" } }` for Claude models when tools are present, ensuring stricter schema adherence.
+- **Adaptive Thinking Budgets**: Replaced static thinking budget values with a dynamic `MODEL_SPECS` mapping. `gemini-3-flash` now correctly uses adaptive thinking budgets (`-1`) which allows the model to decide its own optimal reasoning length, while Pro models use strict budgets (e.g. `10001` for high).
+- **Max Output Tokens Enforcement**: The proxy now enforces hard `maxOutputTokens` caps based on the specific model's upper limits (e.g. `65535` vs `64000`), dynamically adjusting them to ensure there is enough room for both the thinking budget and the final output response without triggering upstream validation errors.
+
+## [1.12.3] - 2026-05-18
+
+### Fixed
+- **Gemini 3.1 Pro High Deprecation (`400 Invalid Argument`)**: Google Cloud Code Assist deprecated the internal string `"gemini-3.1-pro-high"` and replaced it with `"gemini-pro-agent"`. The proxy now automatically maps `"gemini-3.1-pro-high"` to `"gemini-pro-agent"` under the hood when constructing the upstream payload, preventing `400` validation errors while allowing clients to continue using the `-high` alias.
+- **Missing `thought_signature` on Tool Calls (`400 Invalid Argument`)**: Gemini thinking models strictly require a cryptographic Base64 `thought_signature` for all `functionCall` history parts, which the proxy normally caches in RAM. To prevent API rejection on cache misses (e.g. after a proxy restart or when using synthetic tool IDs), the proxy now gracefully collapses the orphaned tool exchange into a neutral user summary (`[Context: The assistant used tools...]`). This preserves the conversation context without triggering the `400` error or teaching the model bad tool-calling formats.
+
 ## [1.12.2] - 2026-05-18
 
 ### Fixed

package/README.md

@@ -15,7 +15,6 @@ Multi-account rotation proxy for Google Antigravity. Distributes API usage acros
 - **Protective pause** -- Pauses all routing for several hours after serious ToS/abuse-style flags so the rest of the pool is not burned
 - **Token auto-refresh** -- Tokens are refreshed automatically before expiry; no manual management
 - **Endpoint cascade** -- Tries daily, autopush, and prod API endpoints for resilience
-- **Pro Family Advisor** -- Scans your account pool and alerts you if there are major imbalances (like some accounts never getting used because of routing bias), giving you actionable steps to optimize token distribution
 - **Advanced Telemetry & Statistics** -- Track exactly how much USD you save compared to a paid API plan, predict quota depletion with the Forecast grid, view Latency tracking (p50/p95), and explore 60-day historical usage heatmaps
 - **Web dashboard** -- Real-time view of model routing table, per-account quota bars with per-model timers, and flagged account alerts
 - **Auto-update notifications** -- The dashboard checks npm for new releases every 30 minutes and shows a banner with one-click update when a newer version is available
@@ -93,13 +92,12 @@ After starting the proxy, open `http://localhost:51200/dashboard` or `http://<yo
 The dashboard shows:
 
 - **Top Status & Controls** -- Real-time routing state, uptime, requests, and PII masking toggle.
-- **Pro Family Advisor & Dual-Window Tracking** -- Advanced logic that tracks and compares both Pro and Free quota windows simultaneously. The Advisor analyzes cumulative quota to suggest mathematical upgrades/downgrades.
 - **Token Usage & Savings** -- Interactive chart (`1h`, `2h`, `4h`, `8h`, `12h`, `1d`, `7d`, `1m`) showing token consumption by model, with estimated USD savings and `CSV`/`JSON` export options.
 - **Activity Heatmap** -- 60-day responsive GitHub-style contribution grid showing request intensity hour by hour.
 - **Latency (p50/p95)** -- Real-time median and 95th percentile tracking for Time-to-First-Byte (TTFB) and Total Duration per model.
 - **Quota Forecast** -- Predictive modeling showing when each model's quota will run out based on the current requests/hour burn rate.
 - **Searchable Request Log** -- Live feed of the last 200 requests with exact timestamps, models, masked accounts, status codes, and latency.
-- **Account Cards** -- Sorted by total quota. Shows status (`active`, `ready`, `cooldown`, `flagged`, `disabled`), dual-window trackers, quota bars with timers, and precise error messages.
+- **Account Cards** -- Sorted by total quota. Shows status (`active`, `ready`, `cooldown`, `flagged`, `disabled`), quota bars with timers, and precise error messages.
 - **Operator Panels** -- "Attention Needed" summaries for quarantined accounts and a real-time event feed of rotator actions.
 
 ![Dashboard](dashboard.png)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-antigravity-rotator",
-  "version": "1.12.2",
+  "version": "1.13.0",
   "description": "Multi-account rotation proxy for Google Antigravity with per-model routing, real-time quota tracking, and infringement detection",
   "license": "MIT",
   "type": "module",

package/src/compat.ts

@@ -86,6 +86,64 @@ export interface CompatCompletion {
 	toolCalls?: OpenAIToolCall[];
 }
 
+// ---------------------------------------------------------------------------
+// Model-specific specs — mirrors Antigravity-Manager model_specs.json
+// ---------------------------------------------------------------------------
+interface ModelSpec {
+	maxOutputTokens: number;
+	thinkingBudget: number; // -1 = adaptive (model decides), >=0 = fixed
+	isThinking: boolean;
+}
+const MODEL_SPECS: Record<string, ModelSpec> = {
+	"gemini-pro-agent":          { maxOutputTokens: 65535, thinkingBudget: 10001, isThinking: true },
+	"gemini-3-flash-agent":      { maxOutputTokens: 65536, thinkingBudget: -1,    isThinking: true },
+	"gemini-3-pro-high":         { maxOutputTokens: 65535, thinkingBudget: 10001, isThinking: true },
+	"gemini-3-pro-low":          { maxOutputTokens: 65535, thinkingBudget: 1001,  isThinking: true },
+	"gemini-3.1-pro-high":       { maxOutputTokens: 65535, thinkingBudget: 10001, isThinking: true },
+	"gemini-3.1-pro-low":        { maxOutputTokens: 65535, thinkingBudget: 1001,  isThinking: true },
+	"gemini-3.1-pro-preview":    { maxOutputTokens: 65535, thinkingBudget: 10001, isThinking: true },
+	"gemini-3-flash":            { maxOutputTokens: 65536, thinkingBudget: 32768, isThinking: true },
+	"gemini-2.5-flash":          { maxOutputTokens: 65535, thinkingBudget: 24576, isThinking: true },
+	"gemini-2.5-pro":            { maxOutputTokens: 65535, thinkingBudget: 1024,  isThinking: true },
+	"claude-sonnet-4-6":         { maxOutputTokens: 64000, thinkingBudget: 32768, isThinking: true },
+	"claude-sonnet-4-6-thinking":{ maxOutputTokens: 64000, thinkingBudget: 32768, isThinking: true },
+	"claude-opus-4-6-thinking":  { maxOutputTokens: 64000, thinkingBudget: 32768, isThinking: true },
+};
+const GEMINI_MAX_OUTPUT_TOKENS = 65536;
+const CLAUDE_MAX_OUTPUT_TOKENS = 64000;
+const FALLBACK_THINKING_BUDGET = 24576;
+const CLAUDE_DEFAULT_THINKING_BUDGET = 32768;
+
+function getModelFamily(model: string): "claude" | "gemini" | "unknown" {
+	const l = model.toLowerCase();
+	if (l.includes("claude")) return "claude";
+	if (l.includes("gemini")) return "gemini";
+	return "unknown";
+}
+
+function getModelSpec(model: string): ModelSpec {
+	const lower = model.toLowerCase();
+	if (MODEL_SPECS[lower]) return MODEL_SPECS[lower];
+	for (const [key, spec] of Object.entries(MODEL_SPECS)) {
+		if (lower.includes(key)) return spec;
+	}
+	const family = getModelFamily(model);
+	if (family === "claude") return { maxOutputTokens: CLAUDE_MAX_OUTPUT_TOKENS, thinkingBudget: CLAUDE_DEFAULT_THINKING_BUDGET, isThinking: true };
+	if (family === "gemini") return { maxOutputTokens: GEMINI_MAX_OUTPUT_TOKENS, thinkingBudget: FALLBACK_THINKING_BUDGET, isThinking: true };
+	return { maxOutputTokens: 65536, thinkingBudget: FALLBACK_THINKING_BUDGET, isThinking: false };
+}
+
+function isThinkingModel(model: string): boolean {
+	const spec = getModelSpec(model);
+	if (spec.isThinking) return true;
+	const l = model.toLowerCase();
+	if (l.includes("gemini")) {
+		const m = l.match(/gemini-(\d+)/);
+		if (m && parseInt(m[1], 10) >= 3) return true;
+	}
+	return false;
+}
+
 type AntigravityPart = { text: string } | { inlineData: { mimeType: string; data: string } };
 
 function isRecord(value: unknown): value is Record<string, unknown> {
@@ -120,12 +178,29 @@ function cacheThoughtSignature(callId: string, signature: string): void {
 	thoughtSignatureCache.set(callId, signature);
 }
 
+/**
+ * Strip cache_control fields from content blocks.
+ * Cloud Code API rejects cache_control with "Extra inputs are not permitted".
+ */
+function cleanCacheControl<T>(content: T): T {
+	if (!Array.isArray(content)) return content;
+	return content.map((block: Record<string, unknown>) => {
+		if (!block || typeof block !== "object") return block;
+		if ("cache_control" in block) {
+			// eslint-disable-next-line @typescript-eslint/no-unused-vars
+			const { cache_control: _cc, ...rest } = block;
+			return rest;
+		}
+		return block;
+	}) as T;
+}
+
 function extractText(content: ChatMessage["content"]): string {
 	if (typeof content === "string") return content;
 	if (!Array.isArray(content)) return "";
-	return content
-		.filter((p) => (p.type === "text" && typeof p.text === "string") || (p.type === "thinking" && typeof p.thinking === "string"))
-		.map((p) => p.type === "thinking" ? `[Thinking]\n${p.thinking}\n[/Thinking]` : p.text)
+	return cleanCacheControl(content)
+		.filter((p: { type?: string; text?: string; thinking?: string }) => (p.type === "text" && typeof p.text === "string") || (p.type === "thinking" && typeof p.thinking === "string"))
+		.map((p: { type?: string; text?: string; thinking?: string }) => p.type === "thinking" ? `[Thinking]\n${p.thinking}\n[/Thinking]` : (p.text as string))
 		.join("\n");
 }
 
@@ -246,12 +321,15 @@ function sanitizeClaudeViaGeminiSchema(schema: unknown): unknown {
 	if (!isRecord(schema)) return schema;
 
 	// Only remove fields that Gemini's API layer truly rejects at the network level.
-	// We keep Draft 2020-12 keywords like minimum/maximum/pattern/title/etc.
+	// We keep standard Draft 2020-12 keywords but must strip exclusiveMinimum/exclusiveMaximum
+	// as boolean values (Draft 4) — the API layer rejects them even for Claude-bound requests.
 	const UNSUPPORTED = new Set([
 		"$schema", "$id", "$ref", "$defs", "definitions",
 		"if", "then", "else", "not",
 		"patternProperties", "unevaluatedProperties", "unevaluatedItems",
 		"contentEncoding", "contentMediaType",
+		// Gemini's protobuf layer rejects these regardless of target model
+		"exclusiveMinimum", "exclusiveMaximum",
 	]);
 
 	const out: Record<string, unknown> = {};
@@ -411,10 +489,41 @@ export function openAIToAntigravityBody(input: OpenAIChatCompletionRequest): Req
 	// Determine if model is Claude — affects schema sanitization and tool call ID handling
 	const isClaude = /^claude-/i.test(input.model);
 
+	// Use model specs to determine thinking support
+	const isThinking = isThinkingModel(input.model);
+	const isGeminiThinking = !isClaude && isThinking;
+
 	const contents: GeminiContent[] = [];
 	for (let i = 0; i < conversationMessages.length; i++) {
 		const msg = conversationMessages[i];
 		if (msg.role === "assistant") {
+			// Check if this is a thinking model turn with tool calls that have no cached signatures.
+			// If so, we collapse the tool exchange into a neutral user summary instead of
+			// injecting [Tool call: ...] text that the model will learn to mimic.
+			const hasMissingSig =
+				isGeminiThinking &&
+				Array.isArray(msg.tool_calls) &&
+				msg.tool_calls.length > 0 &&
+				!thoughtSignatureCache.has(msg.tool_calls[0].id);
+
+			if (hasMissingSig) {
+				// Build a summary of what the model did and what results came back.
+				// We collect the paired tool result(s) from the immediately following messages.
+				const toolNames = msg.tool_calls!.map((tc) => tc.function.name).join(", ");
+				const resultParts: string[] = [];
+				while (i + 1 < conversationMessages.length && conversationMessages[i + 1].role === "tool") {
+					i++;
+					const toolMsg = conversationMessages[i];
+					const toolText = typeof toolMsg.content === "string" ? toolMsg.content : extractText(toolMsg.content);
+					resultParts.push(`${toolMsg.name || "tool"}: ${toolText.slice(0, 500)}`);
+				}
+				const summaryText = `[Context: The assistant used tools (${toolNames}) and received results:\n${resultParts.join("\n")}]`;
+				contents.push({ role: "user", parts: [{ text: summaryText }] });
+				// Add a minimal model acknowledgement to avoid consecutive user turns
+				contents.push({ role: "model", parts: [{ text: "Understood, I have the tool results." }] });
+				continue;
+			}
+
 			const parts: unknown[] = [];
 			if (msg.content) {
 				const textContent = typeof msg.content === "string" ? msg.content : extractText(msg.content);
@@ -427,28 +536,24 @@ export function openAIToAntigravityBody(input: OpenAIChatCompletionRequest): Req
 				// signatures on older historical turns are silently ignored.
 				let isFirstInMessage = true;
 				for (const tc of msg.tool_calls) {
+					let args: unknown;
 					try {
-						const args = typeof tc.function.arguments === "string" ? JSON.parse(tc.function.arguments) : tc.function.arguments;
-						// Only the first functionCall part in a model turn needs the signature
-						const cachedSig = isFirstInMessage ? thoughtSignatureCache.get(tc.id) : undefined;
-						parts.push({
-							...(cachedSig ? { thoughtSignature: cachedSig } : {}),
-							// Include id only for Claude — Gemini native models reject the id field
-							functionCall: { ...(isClaude ? { id: tc.id } : {}), name: tc.function.name, args },
-						});
+						args = typeof tc.function.arguments === "string" ? JSON.parse(tc.function.arguments) : tc.function.arguments;
 					} catch {
-						const cachedSig = isFirstInMessage ? thoughtSignatureCache.get(tc.id) : undefined;
-						parts.push({
-							...(cachedSig ? { thoughtSignature: cachedSig } : {}),
-							functionCall: { ...(isClaude ? { id: tc.id } : {}), name: tc.function.name, args: {} },
-						});
+						args = {};
 					}
+					// Only the first functionCall part in a model turn needs the signature
+					const cachedSig = isFirstInMessage ? thoughtSignatureCache.get(tc.id) : undefined;
+					parts.push({
+						...(cachedSig ? { thoughtSignature: cachedSig } : {}),
+						// Include id only for Claude — Gemini native models reject the id field
+						functionCall: { ...(isClaude ? { id: tc.id } : {}), name: tc.function.name, args },
+					});
 					isFirstInMessage = false;
 				}
 			}
 			if (parts.length > 0) contents.push({ role: "model", parts });
 		} else if (msg.role === "tool") {
-			const prevMsg = conversationMessages[i - 1];
 			const responseText = typeof msg.content === "string" ? msg.content : extractText(msg.content);
 			const fnName = msg.name || "unknown";
 			// Include tool_call_id so Gemini can pass it as tool_use_id to Claude
@@ -460,6 +565,7 @@ export function openAIToAntigravityBody(input: OpenAIChatCompletionRequest): Req
 		} else {
 			// user message
 			const msgParts = extractParts(msg.content);
+
 			if (msgParts.length > 0) contents.push({ role: "user", parts: msgParts });
 		}
 	}
@@ -472,35 +578,84 @@ export function openAIToAntigravityBody(input: OpenAIChatCompletionRequest): Req
 	const geminiTools = convertOpenAIToolsToGemini(inputTools, isClaude);
 	const geminiToolConfig = input.tool_choice !== undefined ? convertToolChoiceToGemini(input.tool_choice) : undefined;
 
-	// Map OpenAI reasoning_effort → Gemini thinkingLevel
-	const thinkingLevel = mapReasoningEffortToThinkingLevel(input.reasoning_effort, input.model);
+	// Cap maxOutputTokens to model limits and build thinkingConfig
+	const modelSpec = getModelSpec(input.model);
+	const modelFamily = getModelFamily(input.model);
+	let maxOutputTokens = typeof input.max_tokens === "number" ? input.max_tokens : undefined;
+	if (maxOutputTokens && maxOutputTokens > modelSpec.maxOutputTokens) {
+		compatLogger.debug(`Capping ${input.model} maxOutputTokens ${maxOutputTokens} → ${modelSpec.maxOutputTokens}`);
+		maxOutputTokens = modelSpec.maxOutputTokens;
+	}
+
+	let thinkingConfigObj: Record<string, unknown> | undefined;
+	if (modelFamily === "claude" && isThinking) {
+		// Claude: snake_case keys required by v1internal
+		const tb = modelSpec.thinkingBudget;
+		thinkingConfigObj = { include_thoughts: true, thinking_budget: tb };
+		if (!maxOutputTokens || maxOutputTokens <= tb) {
+			maxOutputTokens = Math.min(tb + 8192, modelSpec.maxOutputTokens);
+			compatLogger.debug(`Adjusted Claude maxOutputTokens → ${maxOutputTokens}`);
+		}
+	} else if (isThinking) {
+		// Gemini: camelCase keys; thinkingBudget=-1 means adaptive (omit the field)
+		const tb = modelSpec.thinkingBudget;
+		thinkingConfigObj = tb === -1
+			? { includeThoughts: true }
+			: { includeThoughts: true, thinkingBudget: tb };
+		if (tb !== -1 && (!maxOutputTokens || maxOutputTokens <= tb)) {
+			maxOutputTokens = Math.min(tb + 8192, modelSpec.maxOutputTokens);
+			compatLogger.debug(`Adjusted Gemini maxOutputTokens → ${maxOutputTokens}`);
+		}
+	} else if (input.reasoning_effort) {
+		// Non-thinking models with explicit reasoning_effort hint
+		const budgets: Record<string, number> = { low: Math.round(modelSpec.thinkingBudget / 4), medium: Math.round(modelSpec.thinkingBudget / 2), high: modelSpec.thinkingBudget };
+		const b = budgets[input.reasoning_effort.toLowerCase()];
+		if (b) thinkingConfigObj = { includeThoughts: true, thinkingBudget: b };
+	}
+
+	const generationConfig: Record<string, unknown> = {
+		...(typeof input.temperature === "number" ? { temperature: input.temperature } : {}),
+		...(maxOutputTokens ? { maxOutputTokens } : {}),
+		...(thinkingConfigObj ? { thinkingConfig: thinkingConfigObj } : {}),
+	};
 
 	const request: Record<string, unknown> = {
 		contents,
-		generationConfig: {
-			...(typeof input.temperature === "number" ? { temperature: input.temperature } : {}),
-			...(typeof input.max_tokens === "number" ? { maxOutputTokens: input.max_tokens } : {}),
-			// Always request thought blocks. Models that don't support thinking ignore this.
-			thinkingConfig: {
-				includeThoughts: true,
-				...(thinkingLevel ? { thinkingLevel } : {}),
-			},
-		},
+		generationConfig,
 	};
 
 	if (systemParts.length > 0) {
-		request.systemInstruction = {
-			role: "user",
-			parts: [{ text: systemParts.join("\n\n") }],
-		};
+		if (!isClaude && isThinking) {
+			// Gemini thinking models (gemini-3.1-pro-high/low) reject the systemInstruction
+			// field entirely — prepend system prompt to the first user content turn instead.
+			const firstTurn = contents[0];
+			if (firstTurn && firstTurn.role === "user" && (firstTurn.parts[0] as any)?.text !== undefined) {
+				(firstTurn.parts[0] as any).text = systemParts.join("\n\n") + "\n\n" + (firstTurn.parts[0] as any).text;
+			} else if (firstTurn && firstTurn.role === "user") {
+				firstTurn.parts.unshift({ text: systemParts.join("\n\n") + "\n\n" });
+			} else {
+				contents.unshift({
+					role: "user",
+					parts: [{ text: systemParts.join("\n\n") }],
+				});
+			}
+		} else {
+			request.systemInstruction = {
+				role: "system",
+				parts: [{ text: systemParts.join("\n\n") }],
+			};
+		}
 	}
 
 	if (geminiTools.length > 0) request.tools = geminiTools;
 	if (geminiToolConfig) request.toolConfig = geminiToolConfig;
 
+	let mappedModel = input.model;
+	if (mappedModel === "gemini-3.1-pro-high") mappedModel = "gemini-pro-agent";
+
 	return {
 		project: "compat-placeholder",
-		model: input.model,
+		model: mappedModel,
 		userAgent: "antigravity",
 		requestType: "agent",
 		request,
@@ -522,28 +677,47 @@ export function anthropicToAntigravityBody(input: AnthropicMessagesRequest): Req
 }
 
 /**
- * Maps an OpenAI reasoning_effort string to a Gemini thinkingLevel.
- * Gemini 3 Pro only supports LOW and HIGH; Flash supports MINIMAL/LOW/MEDIUM/HIGH.
+ * Maps an OpenAI reasoning_effort / model name suffix to a Gemini thinkingBudget integer.
+ * Cloud Code Assist uses thinkingBudget (integer token count), not thinkingLevel (string).
+ * Values match models.json: -high=10001, -low=1001, flash=dynamic(-1 means dynamic).
+ * Returns undefined for models that don't need an explicit budget (e.g. Claude, plain flash).
  */
-function mapReasoningEffortToThinkingLevel(effort: string | undefined, modelId: string): string | undefined {
-	const isGemini3Pro = /gemini-3(?:\.1)?-pro/i.test(modelId);
-	
+function mapReasoningEffortToThinkingLevel(effort: string | undefined, modelId: string): number | undefined {
+	const lowerModel = modelId.toLowerCase();
+	const isGemini31Pro = /gemini-3\.1-pro/i.test(modelId);
+	const isGemini3Flash = lowerModel.includes("gemini-3-flash");
+
 	let effectiveEffort = effort;
 	if (!effectiveEffort) {
-		const lowerModel = modelId.toLowerCase();
-		if (lowerModel.endsWith("-high") || lowerModel.includes("claude-")) effectiveEffort = "high";
+		if (lowerModel.endsWith("-high") || lowerModel.includes("gemini-pro-agent")) effectiveEffort = "high";
 		else if (lowerModel.endsWith("-low")) effectiveEffort = "low";
-		else if (lowerModel.includes("gemini-3-flash")) effectiveEffort = "high";
+		else if (isGemini3Flash) effectiveEffort = "high";
+		// Claude models: skip — thinking is handled by the anthropic-beta header
 	}
 
 	if (!effectiveEffort) return undefined;
 
-	switch (effectiveEffort.toLowerCase()) {
-		case "low": return isGemini3Pro ? "LOW" : "LOW";
-		case "medium": return isGemini3Pro ? "HIGH" : "MEDIUM";
-		case "high": return "HIGH";
-		default: return undefined;
+	// Gemini 3.1 Pro uses fixed budgets matching models.json
+	if (isGemini31Pro) {
+		switch (effectiveEffort.toLowerCase()) {
+			case "high": return 10001;
+			case "medium": return 5000;
+			case "low": return 1001;
+			default: return undefined;
+		}
+	}
+
+	// Flash uses dynamic budget (-1 means let the model decide)
+	if (isGemini3Flash) {
+		switch (effectiveEffort.toLowerCase()) {
+			case "high": return -1;
+			case "medium": return 4096;
+			case "low": return 1024;
+			default: return undefined;
+		}
 	}
+
+	return undefined;
 }
 
 export function parseAntigravitySse(raw: string): CompatCompletion {
@@ -665,6 +839,10 @@ function writeOpenAIStream(res: ServerResponse, model: string, completion: Compa
 		}
 		res.write(`data: ${JSON.stringify({ id, object: "chat.completion.chunk", created, model, choices: [{ index: 0, delta: {}, finish_reason: "stop" }] })}\n\n`);
 	}
+	// Emit usage chunk so agents (hermes, openwebui) can display token statistics
+	if (completion.inputTokens > 0 || completion.outputTokens > 0) {
+		res.write(`data: ${JSON.stringify({ id, object: "chat.completion.chunk", created, model, choices: [], usage: { prompt_tokens: completion.inputTokens, completion_tokens: completion.outputTokens, total_tokens: completion.inputTokens + completion.outputTokens } })}\n\n`);
+	}
 	res.write("data: [DONE]\n\n");
 	res.end();
 }
@@ -684,7 +862,8 @@ function writeAnthropicStream(res: ServerResponse, model: string, completion: Co
 	res.write(`event: content_block_start\ndata: ${JSON.stringify({ type: "content_block_start", index: contentIndex, content_block: { type: "text", text: "" } })}\n\n`);
 	if (completion.text) res.write(`event: content_block_delta\ndata: ${JSON.stringify({ type: "content_block_delta", index: contentIndex, delta: { type: "text_delta", text: completion.text } })}\n\n`);
 	res.write(`event: content_block_stop\ndata: ${JSON.stringify({ type: "content_block_stop", index: contentIndex })}\n\n`);
-	res.write(`event: message_delta\ndata: ${JSON.stringify({ type: "message_delta", delta: { stop_reason: "end_turn", stop_sequence: null }, usage: { output_tokens: completion.outputTokens } })}\n\n`);
+	// message_delta: include both input_tokens and output_tokens so hermes shows full context count
+	res.write(`event: message_delta\ndata: ${JSON.stringify({ type: "message_delta", delta: { stop_reason: "end_turn", stop_sequence: null }, usage: { input_tokens: completion.inputTokens, output_tokens: completion.outputTokens } })}\n\n`);
 	res.write(`event: message_stop\ndata: ${JSON.stringify({ type: "message_stop" })}\n\n`);
 	res.end();
 }
@@ -812,10 +991,15 @@ async function streamCompatSse(
 	if (!reqClosed && !res.writableEnded) {
 		if (format === "openai") {
 			res.write(`data: ${JSON.stringify({ id, object: "chat.completion.chunk", created, model, choices: [{ index: 0, delta: {}, finish_reason: "stop" }] })}\n\n`);
+			// Emit a usage chunk so agents (hermes, openwebui, etc.) can display token statistics
+			if (inputTokens > 0 || outputTokens > 0) {
+				res.write(`data: ${JSON.stringify({ id, object: "chat.completion.chunk", created, model, choices: [], usage: { prompt_tokens: inputTokens, completion_tokens: outputTokens, total_tokens: inputTokens + outputTokens } })}\n\n`);
+			}
 			res.write("data: [DONE]\n\n");
 		} else {
 			res.write(`event: content_block_stop\ndata: ${JSON.stringify({ type: "content_block_stop", index: 0 })}\n\n`);
-			res.write(`event: message_delta\ndata: ${JSON.stringify({ type: "message_delta", delta: { stop_reason: "end_turn", stop_sequence: null }, usage: { output_tokens: outputTokens } })}\n\n`);
+			// message_delta carries output_tokens; also include input_tokens so Hermes shows full context count
+			res.write(`event: message_delta\ndata: ${JSON.stringify({ type: "message_delta", delta: { stop_reason: "end_turn", stop_sequence: null }, usage: { input_tokens: inputTokens, output_tokens: outputTokens } })}\n\n`);
 			res.write(`event: message_stop\ndata: ${JSON.stringify({ type: "message_stop" })}\n\n`);
 		}
 		res.end();

package/src/dashboard.ts

@@ -519,108 +519,7 @@ const DASHBOARD_HTML = `<!DOCTYPE html>
     font-size: 10px;
     line-height: 1.6;
   }
-  .dw-badge {
-    display: inline-block;
-    width: 32px;
-    text-align: center;
-    font-weight: 700;
-    font-size: 9px;
-    border-radius: 3px;
-    padding: 1px 4px;
-    flex-shrink: 0;
-  }
-  .dw-badge-pro {
-    background: rgba(52, 211, 153, 0.15);
-    color: var(--green);
-  }
-  .dw-badge-free {
-    background: rgba(250, 204, 21, 0.12);
-    color: var(--yellow);
-  }
-  .dw-quota {
-    font-weight: 700;
-    min-width: 28px;
-  }
-  .dw-reset {
-    color: var(--text-dim);
-  }
-  .dw-empty {
-    color: var(--text-dim);
-    font-style: italic;
-    opacity: 0.5;
-  }
-
-  .advisor-panel {
-    background: var(--surface);
-    border: 1px solid var(--border);
-    border-radius: var(--radius);
-    padding: 16px 18px;
-    margin-bottom: 24px;
-  }
-
-  .advisor-title {
-    font-size: 11px;
-    text-transform: uppercase;
-    letter-spacing: 0.8px;
-    color: var(--text-dim);
-    margin-bottom: 10px;
-    display: flex;
-    align-items: center;
-    gap: 8px;
-  }
-
-  .advisor-slots {
-    font-size: 12px;
-    font-family: 'JetBrains Mono', monospace;
-    color: var(--text);
-    margin-left: auto;
-    text-transform: none;
-    letter-spacing: 0;
-  }
-
-  .advisor-action {
-    display: flex;
-    align-items: center;
-    gap: 10px;
-    padding: 8px 10px;
-    margin-bottom: 6px;
-    border-radius: 8px;
-    font-size: 12px;
-  }
-
-  .advisor-action.add-pro {
-    background: rgba(52, 211, 153, 0.06);
-    border-left: 3px solid var(--green);
-  }
-
-  .advisor-action.remove-pro {
-    background: rgba(251, 191, 36, 0.06);
-    border-left: 3px solid var(--yellow);
-  }
-
-  .advisor-action-type {
-    font-weight: 600;
-    font-size: 10px;
-    text-transform: uppercase;
-    letter-spacing: 0.5px;
-    padding: 2px 6px;
-    border-radius: 4px;
-    flex-shrink: 0;
-  }
-
-  .advisor-action.add-pro .advisor-action-type {
-    background: rgba(52, 211, 153, 0.15);
-    color: var(--green);
-  }
-
-  .advisor-action.remove-pro .advisor-action-type {
-    background: rgba(251, 191, 36, 0.15);
-    color: var(--yellow);
-  }
 
-  .advisor-action-label { font-weight: 500; }
-  .advisor-action-reason { color: var(--text-dim); font-size: 11px; margin-left: auto; }
-  .advisor-empty { color: var(--text-dim); font-size: 12px; font-style: italic; }
   .routing-panel {
     border-radius: var(--radius);
     padding: 12px 14px;
@@ -1386,10 +1285,7 @@ const DASHBOARD_HTML = `<!DOCTYPE html>
       <svg viewBox="0 0 24 24"><path d="M12 8v5"/><path d="M12 17.5h.01"/><path d="M10.3 3.8 2.9 17a2 2 0 0 0 1.75 3h14.7A2 2 0 0 0 21.1 17L13.7 3.8a2 2 0 0 0-3.4 0Z"/></svg>
       <span class="header-icon-badge attention" id="attentionBadge" style="display:none">0</span>
     </button>
-    <button class="header-icon-btn advisor" id="advisorBtn" onclick="openModal('advisorModal')" title="Pro Family Advisor" aria-label="Open Pro Family Advisor">
-      <svg viewBox="0 0 24 24"><path d="m5 15 2-9 5 5 5-5 2 9"/><path d="M4 19h16"/></svg>
-      <span class="header-icon-badge advisor" id="advisorBadge" style="display:none">0</span>
-    </button>
+
     <button class="header-icon-btn heart-beat" id="kofiBtn" onclick="openModal('donationModal')" title="Support the Creator" aria-label="Buy me a coffee">
       <svg viewBox="0 0 24 24"><path d="M20.84 4.61a5.5 5.5 0 0 0-7.78 0L12 5.67l-1.06-1.06a5.5 5.5 0 0 0-7.78 7.78l1.06 1.06L12 21.23l7.78-7.78 1.06-1.06a5.5 5.5 0 0 0 0-7.78z"></path></svg>
     </button>
@@ -1481,15 +1377,7 @@ const DASHBOARD_HTML = `<!DOCTYPE html>
   </div>
 </div>
 
-<div class="modal" id="advisorModal" onclick="closeModal(event, 'advisorModal')">
-  <div class="modal-card" onclick="event.stopPropagation()">
-    <div class="modal-header">
-      <strong>Pro Family Advisor</strong>
-      <button class="modal-close" onclick="closeModal(null, 'advisorModal')" aria-label="Close advisor modal">×</button>
-    </div>
-    <div id="proAdvisor"></div>
-  </div>
-</div>
+
 
 <div class="modal" id="donationModal" onclick="closeModal(event, 'donationModal')">
   <div class="modal-card" onclick="event.stopPropagation()" style="max-width: 500px;">
@@ -1582,102 +1470,7 @@ function renderQuotaBars(account) {
   return '<div class="quota-section"><div class="quota-section-title">Quota (per model)</div>' + rows + '</div>';
 }
 
-function renderDualWindows(account) {
-  var qw = account.quotaWindows;
-  if (!qw) return '';
-  var models = Object.keys(qw);
-  if (models.length === 0) return '';
-  var now = Date.now();
-  var rows = models.map(function(modelKey) {
-    var t = qw[modelKey];
-    var shortName = modelKey.split('-').slice(0, 2).join('-');
-    if (shortName === 'claude-opus') shortName = 'claude'; // Clean up Claude display name
-    
-    var proLine = '';
-    var freeLine = '';
-
-    // PRO line
-    if (t.pro && t.pro.lastSeen > 0) {
-      var pQuota = t.pro.lastQuota;
-      var pReset = '';
-      if (t.pro.resetTimeMs > 0) {
-        var pRemain = t.pro.resetTimeMs - now;
-        if (pRemain > 0) {
-          var isRolling5h = pQuota === 100 && Math.abs(pRemain - (5 * 3600000)) < 600000;
-          var isRolling7d = pQuota === 100 && Math.abs(pRemain - (7 * 86400000)) < 600000;
-          if (isRolling5h || isRolling7d) {
-             pReset = '<span style="color:var(--green)">idle</span>';
-          } else {
-             pReset = 'resets in ' + formatDuration(pRemain);
-          }
-        } else {
-          // Reset has passed
-          var was5h = (t.pro.resetTimeMs - t.pro.lastSeen) < (24 * 3600 * 1000);
-          if (was5h) {
-            pQuota = Math.min(100, pQuota + 40);
-            pReset = '<span style="color:var(--green)">+40% idle</span>';
-          } else {
-            pQuota = 100;
-            pReset = '<span style="color:var(--text-dim)">idle</span>';
-          }
-        }
-      }
-      var pqColor = pQuota > 50 ? 'var(--green)' : pQuota > 20 ? 'var(--yellow)' : 'var(--red)';
-      proLine = '<div class="dw-row">' +
-        '<span class="dw-badge dw-badge-pro">PRO</span>' +
-        '<span class="dw-quota" style="color:' + pqColor + '">' + pQuota + '%</span>' +
-        '<span class="dw-reset">' + (pReset || '--') + '</span>' +
-      '</div>';
-    } else {
-      proLine = '<div class="dw-row"><span class="dw-badge dw-badge-pro">PRO</span><span class="dw-empty">no data</span></div>';
-    }
-
-    // FREE line
-    if (t.free && t.free.lastSeen > 0) {
-      var fQuota = t.free.lastQuota;
-      var fReset = '';
-      if (t.free.resetTimeMs > 0) {
-        var fRemain = t.free.resetTimeMs - now;
-        if (fRemain > 0) {
-          var isRolling5h = fQuota === 100 && Math.abs(fRemain - (5 * 3600000)) < 600000;
-          var isRolling7d = fQuota === 100 && Math.abs(fRemain - (7 * 86400000)) < 600000;
-          if (isRolling5h || isRolling7d) {
-             fReset = '<span style="color:var(--green)">idle</span>';
-          } else {
-             fReset = 'resets in ' + formatDuration(fRemain);
-          }
-        } else {
-          // Reset has passed
-          var fWas5h = (t.free.resetTimeMs - t.free.lastSeen) < (24 * 3600 * 1000);
-          if (fWas5h) {
-            fQuota = Math.min(100, fQuota + 40);
-            fReset = '<span style="color:var(--green)">+40% idle</span>';
-          } else {
-            fQuota = 100;
-            fReset = '<span style="color:var(--text-dim)">idle</span>';
-          }
-        }
-      }
-      var fqColor = fQuota > 50 ? 'var(--green)' : fQuota > 20 ? 'var(--yellow)' : 'var(--red)';
-      freeLine = '<div class="dw-row">' +
-        '<span class="dw-badge dw-badge-free">FREE</span>' +
-        '<span class="dw-quota" style="color:' + fqColor + '">' + fQuota + '%</span>' +
-        '<span class="dw-reset">' + (fReset || '--') + '</span>' +
-      '</div>';
-    } else {
-      freeLine = '<div class="dw-row"><span class="dw-badge dw-badge-free">FREE</span><span class="dw-empty">no data</span></div>';
-    }
 
-    return '<div class="dw-model">' +
-      '<div class="dw-model-name">' + shortName + '</div>' +
-      proLine + freeLine +
-    '</div>';
-  }).join('');
-  
-  var swapAllBtn = '<button class="btn-clear-flight" style="margin-left:auto;font-size:8px;padding:1px 4px" title="Manually swap Pro/Free classification for this entire account" onclick="swapWindows(\\'' + jsString(account.email) + '\\')">Swap All</button>';
-  
-  return '<div class="dw-section"><div class="dw-title" style="display:flex;align-items:center">Quota Windows (Pro / Free)' + swapAllBtn + '</div>' + rows + '</div>';
-}
 
 function renderAccounts(data) {
   window.__lastData = data;
@@ -1810,14 +1603,12 @@ function renderAccounts(data) {
         '<div class="card-label">' + escapeHtml(maskText(a.label)) + '</div>' +
         '<div class="card-badges">' +
           (a.proDetected ? '<span class="badge badge-pro">PRO</span>' : '<span class="badge badge-free">FREE</span>') +
-          (a.familyManager ? '<span class="badge badge-fmgr">FAMILY MGR</span>' : '') +
           '<span class="badge badge-' + escapeHtml(a.status) + (isActive ? ' pulse' : '') + '">' + escapeHtml(a.status) + '</span>' +
           modelBadges +
         '</div>' +
       '</div>' +
       '<div class="card-email">' + escapeHtml(maskEmail(a.email)) + '</div>' +
       (a.quota && a.quota.length > 0 ? renderQuotaBars(a) : '') +
-      renderDualWindows(a) +
       '<div class="card-stats">' +
         '<div class="card-stat"><div class="stat-label">Requests</div><div class="stat-value">' +
           a.requestsSinceRotation + ' / ' + a.totalRequests + ' total</div></div>' +
@@ -1851,7 +1642,7 @@ function renderAccounts(data) {
     '</div>';
   }).join('');
 
-  renderProAdvisor(data.proAdvisor);
+
 }
 
 
@@ -1992,7 +1783,6 @@ function renderListView() {
     var tierBadge = a.proDetected
       ? '<span class="badge badge-pro" style="font-size:9px">PRO</span>'
       : '<span class="badge badge-free" style="font-size:9px">FREE</span>';
-    if (a.familyManager) tierBadge += '<span class="badge badge-fmgr" style="font-size:9px">FMGR</span>';
 
     var quotaCell = avgQuota === null
       ? '<span style="color:var(--text-dim)">--</span>'
@@ -2620,6 +2410,7 @@ function renderForecastPanel(data) {
       '<th style="padding:4px 8px">Accounts</th>' +
       '<th style="padding:4px 8px">Burn Rate</th>' +
       '<th style="padding:4px 8px">Estimate</th>' +
+      '<th style="padding:4px 8px">Next Reset</th>' +
     '</tr>';
 
   models.forEach(function(m) {
@@ -2632,6 +2423,23 @@ function renderForecastPanel(data) {
     if (m === 'claude-opus-4-6-thinking') displayName = 'claude';
     if (m === 'gemini-3.1-pro') displayName = 'gemini-3.1-pro';
 
+    var minResetRemaining = null;
+    q.entries.forEach(function(entry) {
+      if (entry.resetTime && entry.timerType !== 'fresh') {
+        var remaining = new Date(entry.resetTime).getTime() - now;
+        if (remaining > 0) {
+          var isRolling5h = entry.percentRemaining === 100 && Math.abs(remaining - (5 * 3600000)) < 600000;
+          var isRolling7d = entry.percentRemaining === 100 && Math.abs(remaining - (7 * 86400000)) < 600000;
+          if (!isRolling5h && !isRolling7d) {
+            if (minResetRemaining === null || remaining < minResetRemaining) {
+              minResetRemaining = remaining;
+            }
+          }
+        }
+      }
+    });
+    var nextResetLabel = minResetRemaining !== null ? formatDuration(minResetRemaining) : '--';
+
     // Estimate: assume ~100 requests per full 100% quota window (empirical)
     // Total remaining "request capacity" ≈ sum of (percent/100 * 100) per account
     var totalCapacity = q.totalPercent; // each 1% ≈ 1 request remaining
@@ -2669,6 +2477,7 @@ function renderForecastPanel(data) {
       '<td style="padding:4px 8px;text-align:center">' + q.accountCount + '</td>' +
       '<td style="padding:4px 8px">' + rateLabel + '</td>' +
       '<td style="padding:4px 8px;color:' + estimateColor + ';font-weight:700">' + estimateLabel + '</td>' +
+      '<td style="padding:4px 8px;color:var(--text-dim)">' + nextResetLabel + '</td>' +
     '</tr>';
   });
 
@@ -2931,44 +2740,7 @@ async function clearCircuitBreaker(modelKey) {
   refresh();
 }
 
-async function swapWindows(email) {
-  if (!confirm('Manually swap Pro and Free data for ALL models on this account? Use this only if the algorithm classified the account tier backward.')) return;
-  await authFetch('/api/account/swap-windows/' + encodeURIComponent(email), { method: 'POST' });
-  refresh();
-}
 
-function renderProAdvisor(advisor) {
-  var panel = document.getElementById('proAdvisor');
-  var button = document.getElementById('advisorBtn');
-  var badge = document.getElementById('advisorBadge');
-  if (!advisor) {
-    panel.innerHTML = '<div class="modal-empty">No advisor data available.</div>';
-    badge.style.display = 'none';
-    button.classList.remove('has-items');
-    return;
-  }
-  var title = '<div class="advisor-title">Pro Family Advisor' +
-    '<span class="advisor-slots">Slots: ' + advisor.currentProCount + '/' + advisor.maxProSlots + '</span></div>';
-  if (advisor.actions.length === 0) {
-    panel.innerHTML = title + '<div class="advisor-empty">No actions recommended</div>';
-    badge.style.display = 'none';
-    button.classList.remove('has-items');
-    return;
-  }
-  var rows = advisor.actions.map(function(a) {
-    var cls = a.type === 'add-pro' ? 'add-pro' : 'remove-pro';
-    var typeLabel = a.type === 'add-pro' ? 'Add Pro' : 'Remove Pro';
-    return '<div class="advisor-action ' + cls + '">' +
-      '<span class="advisor-action-type">' + typeLabel + '</span>' +
-      '<span class="advisor-action-label">' + escapeHtml(maskText(a.label)) + '</span>' +
-      '<span class="advisor-action-reason">' + escapeHtml(a.reason) + '</span>' +
-    '</div>';
-  }).join('');
-  panel.innerHTML = title + rows;
-  badge.style.display = 'inline-flex';
-  badge.textContent = String(advisor.actions.length);
-  button.classList.add('has-items');
-}
 
 function openModal(id) {
   var modal = document.getElementById(id);

package/src/proxy.ts

@@ -1040,26 +1040,7 @@ export function startProxy(rotator: AccountRotator, port: number): void {
 			return;
 		}
 
-		if (method === "POST" && url?.startsWith("/api/account/swap-windows/")) {
-			if (!requireAdmin(req, res)) return;
-			const rest = url.slice("/api/account/swap-windows/".length);
-			const email = decodeURIComponent(rest);
-			const account = rotator.getAccountByEmail(email);
-			if (account && account.quotaWindows) {
-				for (const m of Object.keys(account.quotaWindows)) {
-					const temp = account.quotaWindows[m].pro;
-					account.quotaWindows[m].pro = account.quotaWindows[m].free;
-					account.quotaWindows[m].free = temp;
-				}
-				rotator.saveState();
-				res.writeHead(200);
-				res.end(JSON.stringify({ success: true }));
-			} else {
-				res.writeHead(404);
-				res.end("Account not found");
-			}
-			return;
-		}
+
 
 		if (method === "POST" && (url === "/api/settings/fresh-window-starts/on" || url === "/api/settings/fresh-window-starts/off")) {
 			if (!requireAdmin(req, res)) return;

package/src/rotator.ts

@@ -10,9 +10,6 @@ import {
 	type ModelQuota,
 	type ModelRotationState,
 	type PersistedState,
-	type QuotaWindowHistory,
-	type DualWindowTracker,
-	type ProAdvisorAction,
 	type StatusResponse,
 	type TokenBucket,
 	type TokenUsageData,
@@ -97,7 +94,6 @@ export class AccountRotator {
 			inFlightRequests: 0,
 			inFlightByModel: {},
 			allowFreshWindowStartsOverride: false,
-			quotaWindows: {},
 			dailyRequestCount: 0,
 			dailyRequestDay: currentUtcDay(),
 		}));
@@ -148,7 +144,6 @@ export class AccountRotator {
 					account.disabled = saved.disabled;
 					account.flagged = saved.flagged ?? false;
 				account.allowFreshWindowStartsOverride = saved.allowFreshWindowStartsOverride ?? false;
-					account.quotaWindows = saved.quotaWindows ?? {};
 				}
 			}
 			// Cap any stale cooldowns to 30 min max from now
@@ -230,7 +225,6 @@ export class AccountRotator {
 				disabled: account.disabled,
 				flagged: account.flagged,
 				allowFreshWindowStartsOverride: account.allowFreshWindowStartsOverride,
-					quotaWindows: account.quotaWindows,
 			};
 		}
 			try {
@@ -376,77 +370,6 @@ export class AccountRotator {
 			this.log(`RAW POLL ${account.config.email} -> ${rawLog}`);
 			// ---------------------------------------
 
-			// Record dual-window quota tracking per model (Immutable Anchors Architecture)
-			const now = Date.now();
-			const FIVE_HOURS_10MIN = (5 * 60 + 10) * 60 * 1000;
-			const FIVE_MIN = 5 * 60 * 1000;
-
-			// Step 1: Initialize tracking and check for the definitive PRO signal (genuine 5h timer)
-			let accountIsDefinitivelyPro = false;
-			for (const q of account.quota) {
-				if (!account.quotaWindows[q.modelKey]) {
-					account.quotaWindows[q.modelKey] = {
-						pro: { lastSeen: 0, resetTimeMs: 0, resetTime: null, lastQuota: -1 },
-						free: { lastSeen: 0, resetTimeMs: 0, resetTime: null, lastQuota: -1 },
-					};
-				}
-				if (q.timerType === "5h") {
-					const currentResetMs = q.resetTime ? new Date(q.resetTime).getTime() : 0;
-					if (currentResetMs === 0 || (currentResetMs - now) <= FIVE_HOURS_10MIN) {
-						accountIsDefinitivelyPro = true;
-					}
-				}
-			}
-
-			// Step 2: Update permanent anchors based on the definitive signal
-			for (const q of account.quota) {
-				if (q.timerType === "fresh") continue; // Fresh gives us no reset time to anchor
-				const tracker = account.quotaWindows[q.modelKey];
-				const currentResetMs = q.resetTime ? new Date(q.resetTime).getTime() : 0;
-				if (currentResetMs === 0) continue;
-
-				// Has the real-world time passed the existing Pro anchor?
-				if (tracker.pro.resetTimeMs > 0 && now > tracker.pro.resetTimeMs) {
-					// The old Pro anchor expired naturally. We clear it to make room for a new cycle.
-					tracker.pro.resetTimeMs = 0;
-					tracker.pro.resetTime = null;
-				}
-				// Has the real-world time passed the existing Free anchor?
-				if (tracker.free.resetTimeMs > 0 && now > tracker.free.resetTimeMs) {
-					// The old Free anchor expired naturally. We clear it to make room for a new cycle.
-					tracker.free.resetTimeMs = 0;
-					tracker.free.resetTime = null;
-				}
-
-				const matchesPro = tracker.pro.resetTimeMs > 0 && Math.abs(currentResetMs - tracker.pro.resetTimeMs) < FIVE_MIN;
-				const matchesFree = tracker.free.resetTimeMs > 0 && Math.abs(currentResetMs - tracker.free.resetTimeMs) < FIVE_MIN;
-
-				if (matchesPro) {
-					// It's the Pro window. Update quota.
-					tracker.pro.lastSeen = now;
-					tracker.pro.lastQuota = q.percentRemaining;
-				} else if (matchesFree) {
-					// It's the Free window. Update quota.
-					tracker.free.lastSeen = now;
-					tracker.free.lastQuota = q.percentRemaining;
-				} else {
-					// This is a BRAND NEW reset time (doesn't match either anchor).
-					// We must assign it to either the Pro bucket or the Free bucket.
-					if (accountIsDefinitivelyPro) {
-						// We have absolute proof the account is Pro right now.
-						tracker.pro.lastSeen = now;
-						tracker.pro.resetTimeMs = currentResetMs;
-						tracker.pro.resetTime = q.resetTime;
-						tracker.pro.lastQuota = q.percentRemaining;
-					} else {
-						// We have NO proof the account is Pro. Assume Free.
-						tracker.free.lastSeen = now;
-						tracker.free.resetTimeMs = currentResetMs;
-						tracker.free.resetTime = q.resetTime;
-						tracker.free.lastQuota = q.percentRemaining;
-					}
-				}
-			}
 		} catch {
 			// Network error, skip
 		}
@@ -1468,9 +1391,7 @@ export class AccountRotator {
 					quota: a.quota,
 					inFlightRequests: a.inFlightRequests,
 					inFlightByModel: a.inFlightByModel,
-					proDetected: this.isProAccount(a),
-					quotaWindows: a.quotaWindows,
-				familyManager: !!a.config.familyManager,
+					proDetected: a.config.type === "pro",
 				allowFreshWindowStartsOverride: a.allowFreshWindowStartsOverride,
 				effectiveFreshWindowStartsAllowed: this.isEffectiveFreshWindowAllowed(a),
 			};
@@ -1513,7 +1434,6 @@ export class AccountRotator {
 			},
 			routingHealth,
 			accounts,
-			proAdvisor: this.getProAdvisor(),
 			recentEvents: [...this.recentEvents],
 			requestLog: this.requestLog.slice(0, 100),
 			tokenUsage: this.getTokenUsage(),
@@ -1556,7 +1476,7 @@ export class AccountRotator {
 		).length;
 
 		return {
-			wasProAccount: this.isProAccount(account),
+			wasProAccount: account.config.type === "pro",
 			accountQuotaPercent: quota,
 			timerType,
 			poolSize: this.accounts.length,
@@ -1599,7 +1519,6 @@ export class AccountRotator {
 					inFlightRequests: 0,
 					inFlightByModel: {},
 				allowFreshWindowStartsOverride: false,
-					quotaWindows: {},
 				dailyRequestCount: 0,
 				dailyRequestDay: currentUtcDay(),
 			};
@@ -1642,153 +1561,6 @@ export class AccountRotator {
 		return this.accounts.find((a) => a.config.email === email);
 	}
 
-	// =========================================================================
-	// Pro Family Sharing Advisor
-	// =========================================================================
-
-	// Model keys relevant for Pro advisor decisions (ignore Flash)
-	private static PRO_ADVISOR_MODELS = ["gemini-3.1-pro", "claude-opus-4-6-thinking"];
-
-	/**
-	 * Check if a model's current 7d timer is the Pro cooldown (not Free).
-	 * Uses the dual-window tracker: compares current resetTime against recorded Pro resetTime.
-	 */
-	private isProOriginatedTimer(account: AccountRuntime, modelKey: string): boolean {
-		const tracker = account.quotaWindows[modelKey];
-		if (!tracker || tracker.pro.lastSeen === 0) return false;
-		
-		const currentQuota = account.quota.find(
-			(q) => q.modelKey.includes(modelKey) || modelKey.includes(q.modelKey),
-		);
-		if (!currentQuota || currentQuota.timerType !== "7d") return false;
-		
-		const currentResetMs = currentQuota.resetTime ? new Date(currentQuota.resetTime).getTime() : 0;
-		if (tracker.pro.resetTimeMs === 0 || currentResetMs === 0) return false;
-
-		// Tight 5-min tolerance against permanent anchor
-		return Math.abs(currentResetMs - tracker.pro.resetTimeMs) < 300000;
-	}
-
-	/**
-	 * An account is currently considered "Pro" if, during the very last quota poll,
-	 * its advisor models were tracked in the PRO bucket of the dual-window tracker.
-	 */
-	private isProAccount(account: AccountRuntime): boolean {
-		if (account.lastQuotaPoll === 0) return false;
-		
-		for (const m of AccountRotator.PRO_ADVISOR_MODELS) {
-			const tracker = account.quotaWindows[m];
-			if (!tracker) continue;
-			// If the Pro window was updated exactly during the last poll, it's Pro.
-			// Give a tiny 1s margin for JS execution timing.
-			if (tracker.pro.lastSeen > 0 && Math.abs(tracker.pro.lastSeen - account.lastQuotaPoll) < 1000) {
-				return true;
-			}
-		}
-		return false;
-	}
-
-	/**
-	 * Get the "other" window's quota info for an account/model.
-	 * If currently showing Pro timer → returns Free window data (and vice versa).
-	 */
-	private getAlternateWindow(account: AccountRuntime, modelKey: string): { type: "pro" | "free"; quota: number; resetTimeMs: number; resetTime: string | null } | null {
-		const tracker = account.quotaWindows[modelKey];
-		if (!tracker) return null;
-		const currentQuota = account.quota.find(
-			(q) => q.modelKey.includes(modelKey) || modelKey.includes(q.modelKey),
-		);
-		if (!currentQuota) return null;
-
-		if (this.isProOriginatedTimer(account, modelKey) || currentQuota.timerType === "5h") {
-			// Currently on Pro — return Free window
-			if (tracker.free.lastSeen === 0) return null;
-			return { type: "free", quota: tracker.free.lastQuota, resetTimeMs: tracker.free.resetTimeMs, resetTime: tracker.free.resetTime };
-		} else {
-			// Currently on Free — return Pro window
-			if (tracker.pro.lastSeen === 0) return null;
-			return { type: "pro", quota: tracker.pro.lastQuota, resetTimeMs: tracker.pro.resetTimeMs, resetTime: tracker.pro.resetTime };
-		}
-	}
-
-	private getProAdvisor(): StatusResponse["proAdvisor"] {
-		const maxSlots = this.config.proSlots ?? 6;
-		const proAccounts = this.accounts.filter((a) => !a.disabled && !a.flagged && this.isProAccount(a));
-		const currentProCount = proAccounts.length;
-		const actions: ProAdvisorAction[] = [];
-
-		// Comparative Quota Analysis Logic (Cumulative Score)
-		for (const account of this.accounts) {
-			if (account.disabled || account.flagged) continue;
-
-			let totalProScore = 0;
-			let totalFreeScore = 0;
-			let hasAnyProData = false;
-			let hasAnyFreeData = false;
-
-			for (const modelKey of AccountRotator.PRO_ADVISOR_MODELS) {
-				const tracker = account.quotaWindows[modelKey];
-				if (!tracker) continue;
-				if (tracker.pro.lastSeen > 0) {
-					totalProScore += Math.max(0, tracker.pro.lastQuota);
-					hasAnyProData = true;
-				}
-				if (tracker.free.lastSeen > 0) {
-					totalFreeScore += Math.max(0, tracker.free.lastQuota);
-					hasAnyFreeData = true;
-				}
-			}
-
-			// If a tier has no data at all, its score is effectively 0
-			const effectivePro = hasAnyProData ? totalProScore : 0;
-			const effectiveFree = hasAnyFreeData ? totalFreeScore : 0;
-
-			const isCurrentlyPro = this.isProAccount(account);
-
-			if (isCurrentlyPro) {
-				// Account is currently in PRO tier
-				if (account.config.familyManager) continue; // Never remove FM
-
-				if (effectiveFree > effectivePro) {
-					actions.push({
-						type: "remove-pro",
-						email: account.config.email,
-						label: account.config.label || account.config.email,
-						reason: `Free tier has significantly more combined quota (${effectiveFree}%) than Pro tier (${effectivePro}%). Downgrade to use Free tokens.`,
-					});
-				} else if (effectivePro === 0 && effectiveFree === 0) {
-					actions.push({
-						type: "remove-pro",
-						email: account.config.email,
-						label: account.config.label || account.config.email,
-						reason: `All quota exhausted (0%). Safe to remove from Pro family to free up a slot.`,
-					});
-				}
-			} else {
-				// Account is currently in FREE tier
-				if (effectivePro > effectiveFree) {
-					actions.push({
-						type: "add-pro",
-						email: account.config.email,
-						label: account.config.label || account.config.email,
-						reason: `Pro tier has significantly more combined quota (${effectivePro}%) than Free tier (${effectiveFree}%). Upgrade to use Pro tokens.`,
-						_diff: effectivePro - effectiveFree, // temporary property for sorting
-					} as ProAdvisorAction & { _diff: number });
-				}
-			}
-		}
-
-		// Sort add-pro actions by highest Pro quota difference
-		actions.sort((a, b) => {
-			if (a.type === "add-pro" && b.type === "add-pro") {
-				return ((b as any)._diff || 0) - ((a as any)._diff || 0);
-			}
-			return 0;
-		});
-
-		return { currentProCount, maxProSlots: maxSlots, actions };
-	}
-
 	private shouldUseRequestCountRotation(account: AccountRuntime, model?: string): boolean {
 		if (!this.config.useRequestCountRotationWhenQuotaUnknownOnly) return true;
 		const modelKey = model ? resolveQuotaModelKey(model) : null;

package/src/telemetry.ts

@@ -67,7 +67,6 @@ export function trackFeature(feature: string): void {
 export function getFeaturesSnapshot(): Record<string, boolean> {
 	return {
 		dashboard: _featuresUsed.has("dashboard"),
-		proAdvisor: _featuresUsed.has("proAdvisor"),
 		freshWindowToggle: _featuresUsed.has("freshWindowToggle"),
 		hostedLogin: _featuresUsed.has("hostedLogin"),
 	};

package/src/types.ts

@@ -11,8 +11,6 @@ export interface AccountConfig {
 	label?: string;
 	// Optional - pro/free is detected dynamically from quota API reset times
 	type?: AccountType;
-	// This account owns the family plan and can never be removed from Pro
-	familyManager?: boolean;
 }
 
 export interface Config {
@@ -23,8 +21,6 @@ export interface Config {
 	rotateOnQuotaDrop: number;
 	// How often to poll quota (ms). Default: 5min
 	quotaPollIntervalMs: number;
-	// Max simultaneous Pro accounts (owner + members). Default: 6
-	proSlots?: number;
 	// Hard cap on parallel requests per account. Conservative default is 1.
 	maxConcurrentRequestsPerAccount?: number;
 	// Hard cap on parallel requests per projectId/model. Conservative default is 1.
@@ -159,7 +155,6 @@ export interface AccountRuntime {
 	inFlightRequests: number;
 	inFlightByModel: Record<string, number>;
 	allowFreshWindowStartsOverride: boolean;
-	quotaWindows: QuotaWindowHistory;
 	dailyRequestCount: number;
 	dailyRequestDay: string;
 }
@@ -171,21 +166,6 @@ export interface ModelRotationState {
 	requestsOnActiveAccount: number;
 }
 
-// Persisted state across restarts
-export interface QuotaWindowInfo {
-	lastSeen: number;           // timestamp of last observation
-	resetTimeMs: number;        // epoch ms of the resetTime
-	resetTime: string | null;   // ISO string for display
-	lastQuota: number;          // percentRemaining when last seen
-}
-
-export interface DualWindowTracker {
-	pro: QuotaWindowInfo;
-	free: QuotaWindowInfo;
-}
-
-// Per-account quota window tracking: keyed by model key
-export type QuotaWindowHistory = Record<string, DualWindowTracker>;
 
 export interface PersistedSafetyState {
 	day: string;
@@ -218,7 +198,6 @@ export interface PersistedState {
 			disabled: boolean;
 			flagged: boolean;
 			allowFreshWindowStartsOverride?: boolean;
-			quotaWindows?: QuotaWindowHistory;
 		}
 	>;
 }
@@ -275,12 +254,6 @@ export interface StatusResponse {
 		disabledCount: number;
 		errorCount: number;
 	};
-	// Pro family sharing advisor
-	proAdvisor: {
-		currentProCount: number;
-		maxProSlots: number;
-		actions: ProAdvisorAction[];
-	};
 	recentEvents: RecentEvent[];
 	requestLog: RequestLogEntry[];
 	tokenUsage: TokenUsageData;
@@ -307,19 +280,10 @@ export interface AccountStatus {
 	inFlightByModel: Record<string, number>;
 	// Pro family sharing
 	proDetected: boolean;
-	quotaWindows: QuotaWindowHistory;
-	familyManager: boolean;
 	allowFreshWindowStartsOverride: boolean;
 	effectiveFreshWindowStartsAllowed: boolean;
 }
 
-// Pro advisor suggestion
-export interface ProAdvisorAction {
-	type: "add-pro" | "remove-pro";
-	email: string;
-	label: string;
-	reason: string;
-}
 
 export interface RecentEvent {
 	timestamp: number;