@oh-my-pi/pi-ai 18.3.1 → 18.3.2

package/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 ## [Unreleased]
 
+## [18.3.2] - 2026-09-25
+
+### Fixed
+
+- Fixed capped Anthropic and Bedrock Claude requests with thinking enabled, including on-demand compaction, ending at `max_tokens` with no answer; every capped request now gets its effort's thinking budget on top of the requested output ([#13300](https://github.com/can1357/oh-my-pi/pull/13300) by [@alphastorm](https://github.com/alphastorm))
+
 ## [18.3.1] - 2026-09-25
 
 ### Added

package/dist/types/providers/anthropic-slow-mode.d.ts

@@ -8,6 +8,10 @@
  * capacity right now" with a 429 `slot_busy` or a 529 overload that the client
  * retries after the server-stated interval.
  *
+ * Before the slow lane, the server may grant a small wrap-up allowance (drawn
+ * from the weekly limit) past the session limit: 2xx responses then carry
+ * `anthropic-ratelimit-unified-grace-{5h,7d}-utilization` above zero.
+ *
  * This module owns the wire contract only (header names, parsing). The mode's
  * state machine lives with the caller, which plugs into the Anthropic provider
  * through {@link AnthropicSlowModeHooks}.
@@ -43,6 +47,17 @@ export interface AnthropicSlowModeSignal {
     unifiedLimitClaim: boolean;
     /** True when extra usage (overage) is serving this account. */
     overageInUse: boolean;
+    /**
+     * Wrap-up allowance usage (0..1) past the 5-hour and weekly limits; any
+     * value above zero means the request ran on the allowance. Present only on
+     * responses carrying `anthropic-ratelimit-unified-status`.
+     */
+    graceUtilization?: {
+        fiveHour: number;
+        sevenDay: number;
+    };
+    /** True when `anthropic-ratelimit-unified-overage-status` lets extra usage serve requests. */
+    overageAllowed?: boolean;
 }
 /** One pre-content failure the provider hands to {@link AnthropicSlowModeHooks.onFailure}. */
 export interface AnthropicSlowModeFailure {

package/dist/types/types.d.ts

@@ -443,9 +443,10 @@ export interface StreamOptions {
      */
     fallbackCreditRedemption?: AnthropicFallbackCreditHandle;
     /**
-     * Anthropic subscription slow-mode state machine (Claude Code `/low-priority`).
-     * Consulted only for first-party OAuth `anthropic` requests: stamps
-     * `anthropic-usage-limit: slow` while active and decides capacity waits.
+     * Anthropic subscription usage-limit state machine (wrap-up allowance and
+     * Claude Code's `/low-priority`). Consulted only for first-party OAuth
+     * `anthropic` requests: stamps `anthropic-usage-limit: slow` while active,
+     * observes limit headers, and decides capacity waits.
      */
     anthropicSlowMode?: AnthropicSlowModeHooks;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-ai",
-	"version": "18.3.1",
+	"version": "18.3.2",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"keywords": [
 		"ai",
@@ -155,11 +155,11 @@
 		"fmt": "oxfmt --no-error-on-unmatched-pattern 'src/**/*.{ts,tsx}' '{test,bench,examples,scripts}/**/*.ts' '*.ts'"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.3.1",
-		"@oh-my-pi/pi-catalog": "18.3.1",
-		"@oh-my-pi/pi-natives": "18.3.1",
-		"@oh-my-pi/pi-utils": "18.3.1",
-		"@oh-my-pi/pi-wire": "18.3.1"
+		"@oh-my-pi/omptype": "18.3.2",
+		"@oh-my-pi/pi-catalog": "18.3.2",
+		"@oh-my-pi/pi-natives": "18.3.2",
+		"@oh-my-pi/pi-utils": "18.3.2",
+		"@oh-my-pi/pi-wire": "18.3.2"
 	},
 	"devDependencies": {
 		"@types/bun": "^1.3.14"

package/src/providers/anthropic-slow-mode.ts

@@ -8,6 +8,10 @@
  * capacity right now" with a 429 `slot_busy` or a 529 overload that the client
  * retries after the server-stated interval.
  *
+ * Before the slow lane, the server may grant a small wrap-up allowance (drawn
+ * from the weekly limit) past the session limit: 2xx responses then carry
+ * `anthropic-ratelimit-unified-grace-{5h,7d}-utilization` above zero.
+ *
  * This module owns the wire contract only (header names, parsing). The mode's
  * state machine lives with the caller, which plugs into the Anthropic provider
  * through {@link AnthropicSlowModeHooks}.
@@ -55,6 +59,14 @@ export interface AnthropicSlowModeSignal {
 	unifiedLimitClaim: boolean;
 	/** True when extra usage (overage) is serving this account. */
 	overageInUse: boolean;
+	/**
+	 * Wrap-up allowance usage (0..1) past the 5-hour and weekly limits; any
+	 * value above zero means the request ran on the allowance. Present only on
+	 * responses carrying `anthropic-ratelimit-unified-status`.
+	 */
+	graceUtilization?: { fiveHour: number; sevenDay: number };
+	/** True when `anthropic-ratelimit-unified-overage-status` lets extra usage serve requests. */
+	overageAllowed?: boolean;
 }
 
 /** One pre-content failure the provider hands to {@link AnthropicSlowModeHooks.onFailure}. */
@@ -125,6 +137,12 @@ function readNonNegative(headers: HeadersLike, name: string): number | undefined
 	return Number.isFinite(value) && value >= 0 ? value : undefined;
 }
 
+/** Utilization header clamped to 0..1; absent or malformed reads as 0. */
+function readUtilization(headers: HeadersLike, name: string): number {
+	const value = readNonNegative(headers, name);
+	return value === undefined ? 0 : Math.min(1, value);
+}
+
 function parseStatus(raw: string | undefined): AnthropicSlowStatus | undefined {
 	if (raw === undefined) return undefined;
 	switch (raw.trim()) {
@@ -168,6 +186,15 @@ export function parseAnthropicSlowModeHeaders(headers: HeadersLike): AnthropicSl
 		readHeader(headers, "anthropic-ratelimit-unified-overage-status"),
 	);
 	const overageInUse = readHeader(headers, "anthropic-ratelimit-unified-overage-in-use")?.trim() === "true";
+	const overageStatus = readHeader(headers, "anthropic-ratelimit-unified-overage-status")?.trim();
+	const overageAllowed = overageStatus === "allowed" || overageStatus === "allowed_warning";
+	const graceUtilization =
+		readHeader(headers, "anthropic-ratelimit-unified-status") === undefined
+			? undefined
+			: {
+					fiveHour: readUtilization(headers, "anthropic-ratelimit-unified-grace-5h-utilization"),
+					sevenDay: readUtilization(headers, "anthropic-ratelimit-unified-grace-7d-utilization"),
+				};
 	const signal: AnthropicSlowModeSignal = {
 		...(offer !== undefined ? { offer } : {}),
 		...(status !== undefined ? { status } : {}),
@@ -180,6 +207,8 @@ export function parseAnthropicSlowModeHeaders(headers: HeadersLike): AnthropicSl
 		...(weeklyResetAtSec !== undefined ? { weeklyResetAtSec } : {}),
 		unifiedLimitClaim,
 		overageInUse,
+		...(overageAllowed ? { overageAllowed } : {}),
+		...(graceUtilization !== undefined ? { graceUtilization } : {}),
 	};
 	const hasSlowFacts =
 		offer !== undefined ||
@@ -188,7 +217,15 @@ export function parseAnthropicSlowModeHeaders(headers: HeadersLike): AnthropicSl
 		maxWaitSec !== undefined ||
 		budgetUtilization !== undefined ||
 		budgetResetAtSec !== undefined;
-	if (!hasSlowFacts && !unifiedLimitClaim && fiveHourResetAtSec === undefined && unifiedResetAtSec === undefined) {
+	// A unified-status response always carries wrap-up facts, even when both
+	// grace readings are zero: that is how the controller sees the window close.
+	if (
+		!hasSlowFacts &&
+		graceUtilization === undefined &&
+		!unifiedLimitClaim &&
+		fiveHourResetAtSec === undefined &&
+		unifiedResetAtSec === undefined
+	) {
 		return undefined;
 	}
 	return signal;

package/src/stream.ts

@@ -2138,11 +2138,21 @@ function mapOptionsForApi<TApi extends Api>(
 					? mapEffortToAnthropicAdaptiveEffort(model, reasoning)
 					: undefined;
 
+			// A caller's maxTokens is the output it asked for, but thinking spends the
+			// same max_tokens: adaptive thinking can use all of it and leave no answer.
+			// Give thinking its budget on top, as the budget-only path below does. An
+			// uncapped request keeps the provider default.
+			const maxTokensWithThinking =
+				base.maxTokens === undefined
+					? undefined
+					: maxTokensWithThinkingBudget(base.maxTokens, model.maxTokens, thinkingBudget);
+
 			// For Opus 4.6+ and Sonnet 4.6+: use adaptive thinking with effort level
 			// For older models: use budget-based thinking
 			if (thinkingMode === "anthropic-adaptive") {
 				return castApi<"anthropic-messages">({
 					...base,
+					maxTokens: maxTokensWithThinking,
 					requestModelId: resolveWireModelId(model, reasoning),
 					thinkingEnabled: true,
 					effort,
@@ -2155,6 +2165,7 @@ function mapOptionsForApi<TApi extends Api>(
 			if (ANTHROPIC_USE_INTERLEAVED_THINKING) {
 				return castApi<"anthropic-messages">({
 					...base,
+					maxTokens: maxTokensWithThinking,
 					requestModelId: resolveWireModelId(model, reasoning),
 					thinkingEnabled: true,
 					thinkingBudgetTokens: thinkingBudget,
@@ -2214,8 +2225,25 @@ function mapOptionsForApi<TApi extends Api>(
 				guardrailTrace: model.guardrailTrace ?? options?.guardrailTrace,
 				requestMetadata: options?.requestMetadata,
 			};
-			// Effort modes send effort directly, no budget_tokens — skip budget inflation.
-			if (model.thinking?.mode === "effort" || model.thinking?.mode === "anthropic-adaptive") {
+			// Adaptive Claude shares max_tokens between thinking and the answer, like
+			// the anthropic-messages adaptive path: a caller's cap is the output it
+			// wants, so add the effort's budget on top. Uncapped requests keep the
+			// provider default.
+			if (model.thinking?.mode === "anthropic-adaptive") {
+				const reasoning = bedrockBase.reasoning;
+				const budget = reasoning
+					? (options?.thinkingBudgets?.[reasoning] ?? BEDROCK_CLAUDE_THINKING[reasoning])
+					: 0;
+				if (!model.reasoning || bedrockBase.maxTokens === undefined || budget <= 0) {
+					return castApi<"bedrock-converse-stream">(bedrockBase);
+				}
+				return castApi<"bedrock-converse-stream">({
+					...bedrockBase,
+					maxTokens: maxTokensWithThinkingBudget(bedrockBase.maxTokens, model.maxTokens, budget),
+				});
+			}
+			// Effort mode sends effort directly, no budget_tokens — skip budget inflation.
+			if (model.thinking?.mode === "effort") {
 				return castApi<"bedrock-converse-stream">(bedrockBase);
 			}
 			const budgetInfo = resolveBedrockThinkingBudget(model as Model<"bedrock-converse-stream">, options);

package/src/types.ts

@@ -639,9 +639,10 @@ export interface StreamOptions {
 	 */
 	fallbackCreditRedemption?: AnthropicFallbackCreditHandle;
 	/**
-	 * Anthropic subscription slow-mode state machine (Claude Code `/low-priority`).
-	 * Consulted only for first-party OAuth `anthropic` requests: stamps
-	 * `anthropic-usage-limit: slow` while active and decides capacity waits.
+	 * Anthropic subscription usage-limit state machine (wrap-up allowance and
+	 * Claude Code's `/low-priority`). Consulted only for first-party OAuth
+	 * `anthropic` requests: stamps `anthropic-usage-limit: slow` while active,
+	 * observes limit headers, and decides capacity waits.
 	 */
 	anthropicSlowMode?: AnthropicSlowModeHooks;
 }