llmz 0.9.12 → 0.9.13

package/dist/{common-DJkSpeqw.cjs → common-CaWHSEW4.cjs}

@@ -599,6 +599,13 @@ var StreamingMessageParser = class {
 };
 //#endregion
 //#region src/message-stream/response-parser.ts
+/**
+* Text a model may write before ■start. Some models stream junk for minutes
+* without ever opening the envelope; past this many characters, give up.
+*/
+const MAX_PREAMBLE_LENGTH = 8e3;
+/** Text a model may write after ■end before the rest of its stream is ignored. */
+const MAX_TRAILER_LENGTH = 2e3;
 /** One wire format: a complete ■start … ■end envelope containing valid blocks. */
 var ResponseParser = class {
 	_parser = new StreamingMessageParser();
@@ -611,9 +618,19 @@ var ResponseParser = class {
 	_finished = false;
 	_failure;
 	_trailer = "";
+	_preambleLength = 0;
 	get framed() {
 		return this._mode === "framed";
 	}
+	/** Whether ■end was received. Later text is only kept as a diagnostic. */
+	get closed() {
+		return this._closed;
+	}
+	/** The model is writing junk past either boundary: stop reading the stream. */
+	get done() {
+		if (this._closed) return this._trailer.length > MAX_TRAILER_LENGTH;
+		return this._mode === "detect" && this._preambleLength > 8e3;
+	}
 	get valid() {
 		return this._finished && this._closed && !this._failure;
 	}
@@ -637,7 +654,13 @@ var ResponseParser = class {
 			forward = "";
 		};
 		for (const char of chunk) {
+			if (this._closed) {
+				this._trailer += char;
+				continue;
+			}
+			if (this.done) continue;
 			if (this._mode === "detect") {
+				this._preambleLength++;
 				if (this._discardPrefixLine) {
 					if (char === "\n") this._discardPrefixLine = false;
 					continue;
@@ -668,10 +691,6 @@ var ResponseParser = class {
 				}
 				continue;
 			}
-			if (this._closed) {
-				this._trailer += char;
-				continue;
-			}
 			if (this._boundary || char === "■") {
 				this._boundary += char;
 				if (this._boundary === "■end") {
@@ -693,7 +712,8 @@ var ResponseParser = class {
 		this._finished = true;
 		const events = [];
 		if (this._mode === "detect") {
-			if (this._prefix.trim() === "■start") this._mode = "framed";
+			if (this.done) this._fail(`Missing ■start: no response envelope within ${MAX_PREAMBLE_LENGTH} characters`);
+			else if (this._prefix.trim() === "■start") this._mode = "framed";
 			else this._fail("Missing ■start: every response must begin with the response envelope");
 			this._prefix = "";
 		}

package/dist/{common-DuEhn6u3.js → common-P6vdw8Vv.js}

@@ -599,6 +599,13 @@ var StreamingMessageParser = class {
 };
 //#endregion
 //#region src/message-stream/response-parser.ts
+/**
+* Text a model may write before ■start. Some models stream junk for minutes
+* without ever opening the envelope; past this many characters, give up.
+*/
+const MAX_PREAMBLE_LENGTH = 8e3;
+/** Text a model may write after ■end before the rest of its stream is ignored. */
+const MAX_TRAILER_LENGTH = 2e3;
 /** One wire format: a complete ■start … ■end envelope containing valid blocks. */
 var ResponseParser = class {
 	_parser = new StreamingMessageParser();
@@ -611,9 +618,19 @@ var ResponseParser = class {
 	_finished = false;
 	_failure;
 	_trailer = "";
+	_preambleLength = 0;
 	get framed() {
 		return this._mode === "framed";
 	}
+	/** Whether ■end was received. Later text is only kept as a diagnostic. */
+	get closed() {
+		return this._closed;
+	}
+	/** The model is writing junk past either boundary: stop reading the stream. */
+	get done() {
+		if (this._closed) return this._trailer.length > MAX_TRAILER_LENGTH;
+		return this._mode === "detect" && this._preambleLength > 8e3;
+	}
 	get valid() {
 		return this._finished && this._closed && !this._failure;
 	}
@@ -637,7 +654,13 @@ var ResponseParser = class {
 			forward = "";
 		};
 		for (const char of chunk) {
+			if (this._closed) {
+				this._trailer += char;
+				continue;
+			}
+			if (this.done) continue;
 			if (this._mode === "detect") {
+				this._preambleLength++;
 				if (this._discardPrefixLine) {
 					if (char === "\n") this._discardPrefixLine = false;
 					continue;
@@ -668,10 +691,6 @@ var ResponseParser = class {
 				}
 				continue;
 			}
-			if (this._closed) {
-				this._trailer += char;
-				continue;
-			}
 			if (this._boundary || char === "■") {
 				this._boundary += char;
 				if (this._boundary === "■end") {
@@ -693,7 +712,8 @@ var ResponseParser = class {
 		this._finished = true;
 		const events = [];
 		if (this._mode === "detect") {
-			if (this._prefix.trim() === "■start") this._mode = "framed";
+			if (this.done) this._fail(`Missing ■start: no response envelope within ${MAX_PREAMBLE_LENGTH} characters`);
+			else if (this._prefix.trim() === "■start") this._mode = "framed";
 			else this._fail("Missing ■start: every response must begin with the response envelope");
 			this._prefix = "";
 		}

package/dist/{custom-client-BXhs5PpQ.cjs → custom-client-C3EnKARC.cjs}

@@ -1,4 +1,4 @@
-const require_dual_modes = require("./dual-modes-zxiDUl2A.cjs");
+const require_dual_modes = require("./dual-modes-D3cSQ8j4.cjs");
 const require__overArg = require("./_overArg-C5Hy7oCi.cjs");
 const require__baseIteratee = require("./_baseIteratee-CF4qP2a8.cjs");
 const require_utils = require("./utils-DGQ4sRke.cjs");

package/dist/{custom-client-BvA4375q.js → custom-client-CZVu-wF4.js}

@@ -1,4 +1,4 @@
-import { a as flatRest, i as inspect, o as copyArray, r as extractType, s as baseCreate, t as DualModePrompt } from "./dual-modes-D3_Jtxv1.js";
+import { a as flatRest, i as inspect, o as copyArray, r as extractType, s as baseCreate, t as DualModePrompt } from "./dual-modes-BdZuONje.js";
 import { a as Symbol, n as isArray, o as root, r as isObjectLike } from "./_overArg-CRYbj1NH.js";
 import { D as isBuffer, E as baseUnary, M as baseAssignValue, R as isObject, T as nodeUtil, _ as baseGet, c as _getTag_default, f as getSymbols, h as Stack, k as isPrototype, s as Uint8Array, u as getAllKeys, v as toKey, x as keys, y as castPath, z as arrayMap } from "./_baseIteratee-P_firB7b.js";
 import { _ as getSymbolsIn, b as keysIn, g as getAllKeysIn, i as getErrorMessage, o as getTokenizer, x as assignValue, y as baseSlice } from "./utils-UBeAFeEI.js";

package/dist/{dual-modes-D3_Jtxv1.js → dual-modes-BdZuONje.js}

@@ -1,5 +1,5 @@
 import { n as __exportAll } from "./rolldown-runtime-EAsCxrDo.js";
-import { l as NAME_REGEX, n as parseAssistantResponse, r as replacePlaceholders } from "./common-DuEhn6u3.js";
+import { l as NAME_REGEX, n as parseAssistantResponse, r as replacePlaceholders } from "./common-P6vdw8Vv.js";
 import { a as Symbol, n as isArray, r as isObjectLike } from "./_overArg-CRYbj1NH.js";
 import { A as isArrayLike, B as isSymbol, E as baseUnary, F as WeakMap, L as identity, M as baseAssignValue, O as isArguments, P as defineProperty, R as isObject, _ as baseGet, c as _getTag_default, g as arrayPush, i as mapToArray, m as arrayFilter, t as baseIteratee, x as keys, z as arrayMap } from "./_baseIteratee-P_firB7b.js";
 import { n as noop, t as uniq } from "./uniq-eZWVMz0r.js";
@@ -754,7 +754,7 @@ const renderExamples = async (examples, components, exits) => {
 	if (!examples.length) return "";
 	const registry = new ComponentRegistry(components.map(componentToProtocolDefinition));
 	const exitDefinitions = exits.map(exitToProtocolDefinition);
-	const { parseAssistantResponse } = await import("./common-DuEhn6u3.js").then((n) => n.t);
+	const { parseAssistantResponse } = await import("./common-P6vdw8Vv.js").then((n) => n.t);
 	const render = (output) => {
 		var _response$diagnostics;
 		const response = parseAssistantResponse(`■start\n${output}\n■end`);

package/dist/{dual-modes-zxiDUl2A.cjs → dual-modes-D3cSQ8j4.cjs}

@@ -1,5 +1,5 @@
 const require_compiler = require("./compiler-BJJUmNS9.cjs");
-const require_common = require("./common-DJkSpeqw.cjs");
+const require_common = require("./common-CaWHSEW4.cjs");
 const require__overArg = require("./_overArg-C5Hy7oCi.cjs");
 const require__baseIteratee = require("./_baseIteratee-CF4qP2a8.cjs");
 const require_uniq = require("./uniq-BWs20_o8.cjs");
@@ -755,7 +755,7 @@ const renderExamples = async (examples, components, exits) => {
 	if (!examples.length) return "";
 	const registry = new ComponentRegistry(components.map(componentToProtocolDefinition));
 	const exitDefinitions = exits.map(exitToProtocolDefinition);
-	const { parseAssistantResponse } = await Promise.resolve().then(() => require("./common-DJkSpeqw.cjs")).then((n) => n.common_exports);
+	const { parseAssistantResponse } = await Promise.resolve().then(() => require("./common-CaWHSEW4.cjs")).then((n) => n.common_exports);
 	const render = (output) => {
 		var _response$diagnostics;
 		const response = parseAssistantResponse(`■start\n${output}\n■end`);

package/dist/{execute-BTImw3Hx.cjs → execute-DBMmurHH.cjs}

@@ -1,8 +1,8 @@
 const require_compiler = require("./compiler-BJJUmNS9.cjs");
-const require_dual_modes = require("./dual-modes-zxiDUl2A.cjs");
-const require_common = require("./common-DJkSpeqw.cjs");
+const require_dual_modes = require("./dual-modes-D3cSQ8j4.cjs");
+const require_common = require("./common-CaWHSEW4.cjs");
 const require__baseIteratee = require("./_baseIteratee-CF4qP2a8.cjs");
-const require_custom_client = require("./custom-client-BXhs5PpQ.cjs");
+const require_custom_client = require("./custom-client-C3EnKARC.cjs");
 const require_utils = require("./utils-DGQ4sRke.cjs");
 const require_jsx = require("./jsx-BJ6GWXI2.cjs");
 const require_stack_traces = require("./stack-traces-ntw_qhoy.cjs");
@@ -157,11 +157,44 @@ const RESPONSE_LENGTH_BUFFER = {
 };
 /** Maximum time to wait between two stream chunks before considering the stream stalled. */
 const STREAM_INACTIVITY_TIMEOUT = 18e4;
+/**
+* How long to keep reading after ■end for the provider's final chunk. Models
+* that ignore the stop sequence can otherwise stream junk until the caller's deadline.
+*/
+const STREAM_END_GRACE_PERIOD = 1e3;
 /** A syntactically valid prefix must not execute when generation did not finish successfully. */
 const assertSuccessfulGeneration = (metadata) => {
 	if (metadata.provider === "unknown") throw new require_errors.CognitiveError("LLM generation failed: received error metadata with unknown provider");
 	if (metadata.stopReason === "max_tokens" || metadata.stopReason === "content_filter" || metadata.stopReason === "other") throw new require_errors.CognitiveError(`LLM generation did not complete: stopReason=${metadata.stopReason}`);
 };
+/**
+* Cognitive sends usage only in the final chunk, which is never read when the
+* stream is closed early. Estimate it from the messages sent and the output.
+*/
+const estimateMetadata = (model, messages, output) => {
+	const tokenizer = require_utils.getTokenizer();
+	const countMessage = (message) => {
+		if (typeof message.content === "string") return tokenizer.count(message.content);
+		return (message.content ?? []).map((part) => part.type === "text" ? part.text ?? "" : "").reduce((sum, text) => sum + tokenizer.count(text), 0);
+	};
+	const inputTokens = messages.reduce((total, message) => total + countMessage(message), 0);
+	const outputTokens = tokenizer.count(output);
+	const inputCost = inputTokens * model.input.costPer1MTokens / 1e6;
+	const outputCost = outputTokens * model.output.costPer1MTokens / 1e6;
+	const [provider, ...name] = model.id.split(":");
+	return {
+		provider,
+		model: name.join(":"),
+		usage: {
+			inputTokens,
+			inputCost,
+			outputTokens,
+			outputCost
+		},
+		cost: inputCost + outputCost,
+		stopReason: "stop"
+	};
+};
 const getModelOutputLimit = (inputLength) => clamp(RESPONSE_LENGTH_BUFFER.PERCENTAGE * inputLength, RESPONSE_LENGTH_BUFFER.MIN_TOKENS, RESPONSE_LENGTH_BUFFER.MAX_TOKENS);
 const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, onSend, onSendDelta, onRunStart }) => {
 	const startedAt = Date.now();
@@ -279,16 +312,14 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 			} } : {}
 		}, { signal: streamController.signal });
 		const inactivityTimeout = Math.max(STREAM_INACTIVITY_TIMEOUT, ctx.maxTimeToFirstToken ?? 0);
-		const nextChunk = async () => {
+		/** Resolves to undefined when no chunk arrives within the timeout. */
+		const nextChunk = async (timeout) => {
 			let timer;
-			const stalled = new Promise((_, reject) => {
-				timer = setTimeout(() => {
-					streamController.abort("LLM stream stalled");
-					reject(/* @__PURE__ */ new Error(`LLM stream stalled: no data received for ${inactivityTimeout}ms`));
-				}, inactivityTimeout);
+			const expired = new Promise((resolve) => {
+				timer = setTimeout(() => resolve(void 0), timeout);
 			});
 			try {
-				return await Promise.race([stream.next(), stalled]);
+				return await Promise.race([stream.next(), expired]);
 			} finally {
 				clearTimeout(timer);
 			}
@@ -296,15 +327,22 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 		raw = "";
 		let streamCompleted = false;
 		let accepted = false;
+		let closedAt;
+		let servedModel = model.id;
 		try {
 			while (true) {
 				var _chunk$value, _chunk$value2, _chunk$value3, _chunk$value4;
 				let chunk;
 				try {
-					chunk = await nextChunk();
+					chunk = await nextChunk(closedAt === void 0 ? inactivityTimeout : closedAt + STREAM_END_GRACE_PERIOD - Date.now());
 				} catch (thrown) {
 					throw new require_errors.CognitiveError(`LLM generation failed: ${require_utils.getErrorMessage(thrown)}`);
 				}
+				if (!chunk && closedAt !== void 0) break;
+				if (!chunk) {
+					streamController.abort("LLM stream stalled");
+					throw new require_errors.CognitiveError(`LLM generation failed: LLM stream stalled: no data received for ${inactivityTimeout}ms`);
+				}
 				if (chunk.done) {
 					streamCompleted = true;
 					break;
@@ -323,12 +361,14 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 					completions = [];
 					accepted = false;
 					parser = new require_common.ResponseParser();
+					closedAt = void 0;
 					liveItems.clear();
 					liveContent.clear();
 					codeGenerationTraced = false;
 					runCompleted = false;
 					responseMetadata = void 0;
 					attempt = chunk.value.restart.attempt;
+					servedModel = chunk.value.restart.toModel;
 					await preview({
 						...chunk.value.restart,
 						restart: true,
@@ -345,6 +385,12 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 				timeToFirstToken ??= timeToLastToken;
 				raw += delta;
 				await dispatchSends(parser.push(delta));
+				if (parser.done) break;
+				if (parser.closed) closedAt ??= Date.now();
+			}
+			if (!responseMetadata && !streamCompleted) {
+				const served = models.find((details) => details.id === servedModel) ?? await cognitive.getModelDetails(servedModel).catch(() => model);
+				responseMetadata = estimateMetadata(served, messages, raw);
 			}
 			if (!responseMetadata) throw new require_errors.CognitiveError("LLM streaming completed without metadata");
 			assertSuccessfulGeneration(responseMetadata);

package/dist/{execute-Cr_TyTIE.js → execute-hoM7wiQS.js}

@@ -1,8 +1,8 @@
-import { c as toNumber } from "./dual-modes-D3_Jtxv1.js";
-import { a as ResponseParser, i as toParsedAssistantResponse } from "./common-DuEhn6u3.js";
+import { c as toNumber } from "./dual-modes-BdZuONje.js";
+import { a as ResponseParser, i as toParsedAssistantResponse } from "./common-P6vdw8Vv.js";
 import { n as baseIsEqual } from "./_baseIteratee-P_firB7b.js";
-import { d as Snapshot, f as parseExit, l as PartialExecutionResult, n as Context, s as ErrorExecutionResult, t as _CustomModelClient, u as SuccessExecutionResult } from "./custom-client-BvA4375q.js";
-import { d as stripInvalidIdentifiers, i as getErrorMessage, r as fromJSONSchemaCompat, s as init } from "./utils-UBeAFeEI.js";
+import { d as Snapshot, f as parseExit, l as PartialExecutionResult, n as Context, s as ErrorExecutionResult, t as _CustomModelClient, u as SuccessExecutionResult } from "./custom-client-CZVu-wF4.js";
+import { d as stripInvalidIdentifiers, i as getErrorMessage, o as getTokenizer, r as fromJSONSchemaCompat, s as init } from "./utils-UBeAFeEI.js";
 import { t as createJsxComponent } from "./jsx-CIgdGGVs.js";
 import { t as cleanStackTrace } from "./stack-traces-BIE_5Xwf.js";
 import { a as InvalidCodeError, c as SnapshotSignal, i as CognitiveError, l as ThinkSignal, n as CodeExecutionError, o as LoopExceededError, s as Signals, t as AssignmentError } from "./errors-Cg1G0Awu.js";
@@ -156,11 +156,44 @@ const RESPONSE_LENGTH_BUFFER = {
 };
 /** Maximum time to wait between two stream chunks before considering the stream stalled. */
 const STREAM_INACTIVITY_TIMEOUT = 18e4;
+/**
+* How long to keep reading after ■end for the provider's final chunk. Models
+* that ignore the stop sequence can otherwise stream junk until the caller's deadline.
+*/
+const STREAM_END_GRACE_PERIOD = 1e3;
 /** A syntactically valid prefix must not execute when generation did not finish successfully. */
 const assertSuccessfulGeneration = (metadata) => {
 	if (metadata.provider === "unknown") throw new CognitiveError("LLM generation failed: received error metadata with unknown provider");
 	if (metadata.stopReason === "max_tokens" || metadata.stopReason === "content_filter" || metadata.stopReason === "other") throw new CognitiveError(`LLM generation did not complete: stopReason=${metadata.stopReason}`);
 };
+/**
+* Cognitive sends usage only in the final chunk, which is never read when the
+* stream is closed early. Estimate it from the messages sent and the output.
+*/
+const estimateMetadata = (model, messages, output) => {
+	const tokenizer = getTokenizer();
+	const countMessage = (message) => {
+		if (typeof message.content === "string") return tokenizer.count(message.content);
+		return (message.content ?? []).map((part) => part.type === "text" ? part.text ?? "" : "").reduce((sum, text) => sum + tokenizer.count(text), 0);
+	};
+	const inputTokens = messages.reduce((total, message) => total + countMessage(message), 0);
+	const outputTokens = tokenizer.count(output);
+	const inputCost = inputTokens * model.input.costPer1MTokens / 1e6;
+	const outputCost = outputTokens * model.output.costPer1MTokens / 1e6;
+	const [provider, ...name] = model.id.split(":");
+	return {
+		provider,
+		model: name.join(":"),
+		usage: {
+			inputTokens,
+			inputCost,
+			outputTokens,
+			outputCost
+		},
+		cost: inputCost + outputCost,
+		stopReason: "stop"
+	};
+};
 const getModelOutputLimit = (inputLength) => clamp(RESPONSE_LENGTH_BUFFER.PERCENTAGE * inputLength, RESPONSE_LENGTH_BUFFER.MIN_TOKENS, RESPONSE_LENGTH_BUFFER.MAX_TOKENS);
 const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, onSend, onSendDelta, onRunStart }) => {
 	const startedAt = Date.now();
@@ -278,16 +311,14 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 			} } : {}
 		}, { signal: streamController.signal });
 		const inactivityTimeout = Math.max(STREAM_INACTIVITY_TIMEOUT, ctx.maxTimeToFirstToken ?? 0);
-		const nextChunk = async () => {
+		/** Resolves to undefined when no chunk arrives within the timeout. */
+		const nextChunk = async (timeout) => {
 			let timer;
-			const stalled = new Promise((_, reject) => {
-				timer = setTimeout(() => {
-					streamController.abort("LLM stream stalled");
-					reject(/* @__PURE__ */ new Error(`LLM stream stalled: no data received for ${inactivityTimeout}ms`));
-				}, inactivityTimeout);
+			const expired = new Promise((resolve) => {
+				timer = setTimeout(() => resolve(void 0), timeout);
 			});
 			try {
-				return await Promise.race([stream.next(), stalled]);
+				return await Promise.race([stream.next(), expired]);
 			} finally {
 				clearTimeout(timer);
 			}
@@ -295,15 +326,22 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 		raw = "";
 		let streamCompleted = false;
 		let accepted = false;
+		let closedAt;
+		let servedModel = model.id;
 		try {
 			while (true) {
 				var _chunk$value, _chunk$value2, _chunk$value3, _chunk$value4;
 				let chunk;
 				try {
-					chunk = await nextChunk();
+					chunk = await nextChunk(closedAt === void 0 ? inactivityTimeout : closedAt + STREAM_END_GRACE_PERIOD - Date.now());
 				} catch (thrown) {
 					throw new CognitiveError(`LLM generation failed: ${getErrorMessage(thrown)}`);
 				}
+				if (!chunk && closedAt !== void 0) break;
+				if (!chunk) {
+					streamController.abort("LLM stream stalled");
+					throw new CognitiveError(`LLM generation failed: LLM stream stalled: no data received for ${inactivityTimeout}ms`);
+				}
 				if (chunk.done) {
 					streamCompleted = true;
 					break;
@@ -322,12 +360,14 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 					completions = [];
 					accepted = false;
 					parser = new ResponseParser();
+					closedAt = void 0;
 					liveItems.clear();
 					liveContent.clear();
 					codeGenerationTraced = false;
 					runCompleted = false;
 					responseMetadata = void 0;
 					attempt = chunk.value.restart.attempt;
+					servedModel = chunk.value.restart.toModel;
 					await preview({
 						...chunk.value.restart,
 						restart: true,
@@ -344,6 +384,12 @@ const generateCode = async ({ iteration, ctx, cognitive, controller, metadata, o
 				timeToFirstToken ??= timeToLastToken;
 				raw += delta;
 				await dispatchSends(parser.push(delta));
+				if (parser.done) break;
+				if (parser.closed) closedAt ??= Date.now();
+			}
+			if (!responseMetadata && !streamCompleted) {
+				const served = models.find((details) => details.id === servedModel) ?? await cognitive.getModelDetails(servedModel).catch(() => model);
+				responseMetadata = estimateMetadata(served, messages, raw);
 			}
 			if (!responseMetadata) throw new CognitiveError("LLM streaming completed without metadata");
 			assertSuccessfulGeneration(responseMetadata);

package/dist/index.cjs

@@ -1,7 +1,7 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
-const require_dual_modes = require("./dual-modes-zxiDUl2A.cjs");
-const require_common = require("./common-DJkSpeqw.cjs");
-const require_custom_client = require("./custom-client-BXhs5PpQ.cjs");
+const require_dual_modes = require("./dual-modes-D3cSQ8j4.cjs");
+const require_common = require("./common-CaWHSEW4.cjs");
+const require_custom_client = require("./custom-client-C3EnKARC.cjs");
 const require_utils = require("./utils-DGQ4sRke.cjs");
 const require_tool = require("./tool-ikMsh-TB.cjs");
 const require_errors = require("./errors-C962uJdS.cjs");
@@ -12,7 +12,7 @@ const require_truncator = require("./truncator-DmCu9M4m.cjs");
 const require_quickjs_variant = require("./quickjs-variant-CsIwTVFC.cjs");
 let _bpinternal_zui = require("@bpinternal/zui");
 //#region package.json
-var version = "0.9.12";
+var version = "0.9.13";
 //#endregion
 //#region src/hoist.ts
 /**
@@ -1693,7 +1693,7 @@ const utils = {
 * })
 */
 const execute = async (props) => {
-	const { executeContext } = await Promise.resolve().then(() => require("./execute-BTImw3Hx.cjs"));
+	const { executeContext } = await Promise.resolve().then(() => require("./execute-DBMmurHH.cjs"));
 	return executeContext(props);
 };
 /**
@@ -1702,7 +1702,7 @@ const execute = async (props) => {
 * It's recommended to call this function at the beginning of your application without awaiting it (void init())
 */
 const init = async () => {
-	await Promise.resolve().then(() => require("./execute-BTImw3Hx.cjs"));
+	await Promise.resolve().then(() => require("./execute-DBMmurHH.cjs"));
 	await Promise.resolve().then(() => require("./component-CPKXe6Ca.cjs")).then((n) => n.component_exports);
 	await Promise.resolve().then(() => require("./tool-ikMsh-TB.cjs")).then((n) => n.tool_exports);
 	await Promise.resolve().then(() => require("./exit-DTDB1SJ3.cjs")).then((n) => n.exit_exports);
@@ -1711,7 +1711,7 @@ const init = async () => {
 	await Promise.resolve().then(() => require("./utils-DGQ4sRke.cjs")).then((n) => n.utils_exports);
 	await Promise.resolve().then(() => require("./truncator-DmCu9M4m.cjs")).then((n) => n.truncator_exports);
 	await Promise.resolve().then(() => require("./typings-DXJMu59O.cjs")).then((n) => n.typings_exports);
-	await Promise.resolve().then(() => require("./dual-modes-zxiDUl2A.cjs")).then((n) => n.dual_modes_exports);
+	await Promise.resolve().then(() => require("./dual-modes-D3cSQ8j4.cjs")).then((n) => n.dual_modes_exports);
 };
 //#endregion
 exports.BaseResponse = BaseResponse;

package/dist/index.js

@@ -1,6 +1,6 @@
-import { d as validateComponent, f as validateProps, l as Example, p as ComponentRegistry, u as generateInstructions } from "./dual-modes-D3_Jtxv1.js";
-import { c as MARKER, l as NAME_REGEX, o as StreamingMessageParser, s as tryParseJson } from "./common-DuEhn6u3.js";
-import { a as ThinkExit, c as ExecutionResult, d as Snapshot, f as parseExit, i as ListenExit, l as PartialExecutionResult, o as getValue, r as DefaultExit, s as ErrorExecutionResult, t as _CustomModelClient, u as SuccessExecutionResult } from "./custom-client-BvA4375q.js";
+import { d as validateComponent, f as validateProps, l as Example, p as ComponentRegistry, u as generateInstructions } from "./dual-modes-BdZuONje.js";
+import { c as MARKER, l as NAME_REGEX, o as StreamingMessageParser, s as tryParseJson } from "./common-P6vdw8Vv.js";
+import { a as ThinkExit, c as ExecutionResult, d as Snapshot, f as parseExit, i as ListenExit, l as PartialExecutionResult, o as getValue, r as DefaultExit, s as ErrorExecutionResult, t as _CustomModelClient, u as SuccessExecutionResult } from "./custom-client-CZVu-wF4.js";
 import { a as getMultilineComment, c as isJsonSchema, l as isValidIdentifier, m as toValidObjectName, n as escapeString, p as toValidFunctionName, t as configureTokenizer, u as isZuiSchema, v as upperFirst } from "./utils-UBeAFeEI.js";
 import { t as Tool } from "./tool-BoXBgZeM.js";
 import { c as SnapshotSignal, l as ThinkSignal, o as LoopExceededError, r as CodeFormattingError } from "./errors-Cg1G0Awu.js";
@@ -11,7 +11,7 @@ import { i as wrapContent, n as truncateWrappedContent, t as stripTruncationTags
 import { t as configureQuickJS } from "./quickjs-variant-BYhMBpwU.js";
 import { transforms, z } from "@bpinternal/zui";
 //#region package.json
-var version = "0.9.12";
+var version = "0.9.13";
 //#endregion
 //#region src/hoist.ts
 /**
@@ -1692,7 +1692,7 @@ const utils = {
 * })
 */
 const execute = async (props) => {
-	const { executeContext } = await import("./execute-Cr_TyTIE.js");
+	const { executeContext } = await import("./execute-hoM7wiQS.js");
 	return executeContext(props);
 };
 /**
@@ -1701,7 +1701,7 @@ const execute = async (props) => {
 * It's recommended to call this function at the beginning of your application without awaiting it (void init())
 */
 const init = async () => {
-	await import("./execute-Cr_TyTIE.js");
+	await import("./execute-hoM7wiQS.js");
 	await import("./component-DGG7d_Mz.js").then((n) => n.r);
 	await import("./tool-BoXBgZeM.js").then((n) => n.n);
 	await import("./exit-AwQk1cyW.js").then((n) => n.n);
@@ -1710,7 +1710,7 @@ const init = async () => {
 	await import("./utils-UBeAFeEI.js").then((n) => n.h);
 	await import("./truncator-CD7QtE84.js").then((n) => n.r);
 	await import("./typings-CrBa7Tyy.js").then((n) => n.n);
-	await import("./dual-modes-D3_Jtxv1.js").then((n) => n.n);
+	await import("./dual-modes-BdZuONje.js").then((n) => n.n);
 };
 //#endregion
 export { BaseResponse, Chat, CitationsManager, Component, ComponentRegistry, DefaultComponents, DefaultExit, ErrorExecutionResult, Example, ExecutionResult, Exit, ListenExit, LoopExceededError, MARKER, NAME_REGEX, NextResponse, ObjectInstance, PartialExecutionResult, ReplayableAsyncIterable, RunResponse, SendResponse, Snapshot, SnapshotSignal, StreamingMessageParser, SuccessExecutionResult, ThinkExit, ThinkSignal, Tool, _CustomModelClient, assertValidComponent, collectText, configureQuickJS, configureTokenizer, execute, generateInstructions, getValue, init, isAnyComponent, isComponent, normalizeComponentDefinition, normalizeComponentName, normalizePropsSchema, parseExit, parseStream, parseText, sanitizeMessageText, tryParseJson, utils, validateComponent, validateProps, version };

package/dist/index.workerd.js

@@ -52,7 +52,7 @@ var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__ge
 }) : target, mod));
 //#endregion
 //#region package.json
-var version = "0.9.12", quoteExample, quotePartialExample, exampleBoundaryInstructions, quoteResponseExample;
+var version = "0.9.13", quoteExample, quotePartialExample, exampleBoundaryInstructions, quoteResponseExample;
 var init_example_format = __esmMin((() => {
 	quoteExample = (output) => `"""\n${output}\n"""`;
 	quotePartialExample = (output) => quoteExample(`(...)\n${output}\n(...)`);
@@ -1746,10 +1746,12 @@ var init_parser = __esmMin((() => {
 }));
 //#endregion
 //#region src/message-stream/response-parser.ts
-var ResponseParser;
+var MAX_PREAMBLE_LENGTH, MAX_TRAILER_LENGTH, ResponseParser;
 var init_response_parser = __esmMin((() => {
 	init_compiler();
 	init_parser();
+	MAX_PREAMBLE_LENGTH = 8e3;
+	MAX_TRAILER_LENGTH = 2e3;
 	ResponseParser = class {
 		_parser = new StreamingMessageParser();
 		_mode = "detect";
@@ -1761,9 +1763,19 @@ var init_response_parser = __esmMin((() => {
 		_finished = false;
 		_failure;
 		_trailer = "";
+		_preambleLength = 0;
 		get framed() {
 			return this._mode === "framed";
 		}
+		/** Whether ■end was received. Later text is only kept as a diagnostic. */
+		get closed() {
+			return this._closed;
+		}
+		/** The model is writing junk past either boundary: stop reading the stream. */
+		get done() {
+			if (this._closed) return this._trailer.length > MAX_TRAILER_LENGTH;
+			return this._mode === "detect" && this._preambleLength > 8e3;
+		}
 		get valid() {
 			return this._finished && this._closed && !this._failure;
 		}
@@ -1787,7 +1799,13 @@ var init_response_parser = __esmMin((() => {
 				forward = "";
 			};
 			for (const char of chunk) {
+				if (this._closed) {
+					this._trailer += char;
+					continue;
+				}
+				if (this.done) continue;
 				if (this._mode === "detect") {
+					this._preambleLength++;
 					if (this._discardPrefixLine) {
 						if (char === "\n") this._discardPrefixLine = false;
 						continue;
@@ -1818,10 +1836,6 @@ var init_response_parser = __esmMin((() => {
 					}
 					continue;
 				}
-				if (this._closed) {
-					this._trailer += char;
-					continue;
-				}
 				if (this._boundary || char === "■") {
 					this._boundary += char;
 					if (this._boundary === "■end") {
@@ -1843,7 +1857,8 @@ var init_response_parser = __esmMin((() => {
 			this._finished = true;
 			const events = [];
 			if (this._mode === "detect") {
-				if (this._prefix.trim() === "■start") this._mode = "framed";
+				if (this.done) this._fail(`Missing ■start: no response envelope within ${MAX_PREAMBLE_LENGTH} characters`);
+				else if (this._prefix.trim() === "■start") this._mode = "framed";
 				else this._fail("Missing ■start: every response must begin with the response envelope");
 				this._prefix = "";
 			}
@@ -18113,7 +18128,7 @@ var init_vm = __esmMin((() => {
 }));
 //#endregion
 //#region src/runtime/generate.ts
-var RESPONSE_LENGTH_BUFFER, STREAM_INACTIVITY_TIMEOUT, assertSuccessfulGeneration, getModelOutputLimit, generateCode;
+var RESPONSE_LENGTH_BUFFER, STREAM_INACTIVITY_TIMEOUT, STREAM_END_GRACE_PERIOD, assertSuccessfulGeneration, estimateMetadata, getModelOutputLimit, generateCode;
 var init_generate = __esmMin((() => {
 	init_lodash();
 	init_abort_signal();
@@ -18128,10 +18143,35 @@ var init_generate = __esmMin((() => {
 		PERCENTAGE: .1
 	};
 	STREAM_INACTIVITY_TIMEOUT = 18e4;
+	STREAM_END_GRACE_PERIOD = 1e3;
 	assertSuccessfulGeneration = (metadata) => {
 		if (metadata.provider === "unknown") throw new CognitiveError("LLM generation failed: received error metadata with unknown provider");
 		if (metadata.stopReason === "max_tokens" || metadata.stopReason === "content_filter" || metadata.stopReason === "other") throw new CognitiveError(`LLM generation did not complete: stopReason=${metadata.stopReason}`);
 	};
+	estimateMetadata = (model, messages, output) => {
+		const tokenizer = getTokenizer();
+		const countMessage = (message) => {
+			if (typeof message.content === "string") return tokenizer.count(message.content);
+			return (message.content ?? []).map((part) => part.type === "text" ? part.text ?? "" : "").reduce((sum, text) => sum + tokenizer.count(text), 0);
+		};
+		const inputTokens = messages.reduce((total, message) => total + countMessage(message), 0);
+		const outputTokens = tokenizer.count(output);
+		const inputCost = inputTokens * model.input.costPer1MTokens / 1e6;
+		const outputCost = outputTokens * model.output.costPer1MTokens / 1e6;
+		const [provider, ...name] = model.id.split(":");
+		return {
+			provider,
+			model: name.join(":"),
+			usage: {
+				inputTokens,
+				inputCost,
+				outputTokens,
+				outputCost
+			},
+			cost: inputCost + outputCost,
+			stopReason: "stop"
+		};
+	};
 	getModelOutputLimit = (inputLength) => clamp(RESPONSE_LENGTH_BUFFER.PERCENTAGE * inputLength, RESPONSE_LENGTH_BUFFER.MIN_TOKENS, RESPONSE_LENGTH_BUFFER.MAX_TOKENS);
 	generateCode = async ({ iteration, ctx, cognitive, controller, metadata, onSend, onSendDelta, onRunStart }) => {
 		const startedAt = Date.now();
@@ -18249,16 +18289,14 @@ var init_generate = __esmMin((() => {
 				} } : {}
 			}, { signal: streamController.signal });
 			const inactivityTimeout = Math.max(STREAM_INACTIVITY_TIMEOUT, ctx.maxTimeToFirstToken ?? 0);
-			const nextChunk = async () => {
+			/** Resolves to undefined when no chunk arrives within the timeout. */
+			const nextChunk = async (timeout) => {
 				let timer;
-				const stalled = new Promise((_, reject) => {
-					timer = setTimeout(() => {
-						streamController.abort("LLM stream stalled");
-						reject(/* @__PURE__ */ new Error(`LLM stream stalled: no data received for ${inactivityTimeout}ms`));
-					}, inactivityTimeout);
+				const expired = new Promise((resolve) => {
+					timer = setTimeout(() => resolve(void 0), timeout);
 				});
 				try {
-					return await Promise.race([stream.next(), stalled]);
+					return await Promise.race([stream.next(), expired]);
 				} finally {
 					clearTimeout(timer);
 				}
@@ -18266,14 +18304,21 @@ var init_generate = __esmMin((() => {
 			raw = "";
 			let streamCompleted = false;
 			let accepted = false;
+			let closedAt;
+			let servedModel = model.id;
 			try {
 				while (true) {
 					let chunk;
 					try {
-						chunk = await nextChunk();
+						chunk = await nextChunk(closedAt === void 0 ? inactivityTimeout : closedAt + STREAM_END_GRACE_PERIOD - Date.now());
 					} catch (thrown) {
 						throw new CognitiveError(`LLM generation failed: ${getErrorMessage(thrown)}`);
 					}
+					if (!chunk && closedAt !== void 0) break;
+					if (!chunk) {
+						streamController.abort("LLM stream stalled");
+						throw new CognitiveError(`LLM generation failed: LLM stream stalled: no data received for ${inactivityTimeout}ms`);
+					}
 					if (chunk.done) {
 						streamCompleted = true;
 						break;
@@ -18292,12 +18337,14 @@ var init_generate = __esmMin((() => {
 						completions = [];
 						accepted = false;
 						parser = new ResponseParser();
+						closedAt = void 0;
 						liveItems.clear();
 						liveContent.clear();
 						codeGenerationTraced = false;
 						runCompleted = false;
 						responseMetadata = void 0;
 						attempt = chunk.value.restart.attempt;
+						servedModel = chunk.value.restart.toModel;
 						await preview({
 							...chunk.value.restart,
 							restart: true,
@@ -18314,6 +18361,12 @@ var init_generate = __esmMin((() => {
 					timeToFirstToken ??= timeToLastToken;
 					raw += delta;
 					await dispatchSends(parser.push(delta));
+					if (parser.done) break;
+					if (parser.closed) closedAt ??= Date.now();
+				}
+				if (!responseMetadata && !streamCompleted) {
+					const served = models.find((details) => details.id === servedModel) ?? await cognitive.getModelDetails(servedModel).catch(() => model);
+					responseMetadata = estimateMetadata(served, messages, raw);
 				}
 				if (!responseMetadata) throw new CognitiveError("LLM streaming completed without metadata");
 				assertSuccessfulGeneration(responseMetadata);

package/dist/message-stream/response-parser.d.ts

@@ -1,5 +1,12 @@
 import type { StopReason } from '@botpress/cognitive';
 import type { Diagnostic, MessageStreamEvent, ParsedItem } from './types.js';
+/**
+ * Text a model may write before ■start. Some models stream junk for minutes
+ * without ever opening the envelope; past this many characters, give up.
+ */
+export declare const MAX_PREAMBLE_LENGTH = 8000;
+/** Text a model may write after ■end before the rest of its stream is ignored. */
+export declare const MAX_TRAILER_LENGTH = 2000;
 /** One wire format: a complete ■start … ■end envelope containing valid blocks. */
 export declare class ResponseParser {
     private _parser;
@@ -12,7 +19,12 @@ export declare class ResponseParser {
     private _finished;
     private _failure;
     private _trailer;
+    private _preambleLength;
     get framed(): boolean;
+    /** Whether ■end was received. Later text is only kept as a diagnostic. */
+    get closed(): boolean;
+    /** The model is writing junk past either boundary: stop reading the stream. */
+    get done(): boolean;
     get valid(): boolean;
     get items(): ParsedItem[];
     get diagnostics(): Diagnostic[];

package/package.json

@@ -2,7 +2,7 @@
   "name": "llmz",
   "type": "module",
   "description": "LLMz - An LLM-native Typescript VM built on top of Zui",
-  "version": "0.9.12",
+  "version": "0.9.13",
   "files": [
     "dist",
     "DOCS.md"