autotel 3.7.0 → 4.0.0

package/dist/index.cjs

@@ -19,14 +19,13 @@ const require_parse_error = require('./parse-error.cjs');
 const require_drain_pipeline = require('./drain-pipeline.cjs');
 const require_metric_helpers = require('./metric-helpers.cjs');
 const require_semantic_helpers = require('./semantic-helpers.cjs');
-const require_registry = require('./registry-JZg2J3RZ.cjs');
+const require_registry = require('./registry-DYgvb62e.cjs');
 const require_semantic_conventions = require('./semantic-conventions.cjs');
-const require_attributes = require('./attributes-D3etyRVc.cjs');
+const require_attributes = require('./attributes-PZ5doLgw.cjs');
 const require_messaging = require('./messaging.cjs');
 const require_business_baggage = require('./business-baggage.cjs');
 const require_workflow = require('./workflow.cjs');
 let _opentelemetry_api = require("@opentelemetry/api");
-let _opentelemetry_sdk_metrics = require("@opentelemetry/sdk-metrics");
 let node_async_hooks = require("node:async_hooks");
 
 //#region src/trace-hybrid.ts
@@ -643,334 +642,6 @@ function defineEnricher(def, options = {}) {
 	};
 }
 
-//#endregion
-//#region src/gen-ai-metrics.ts
-/**
-* LLM-tuned histogram buckets.
-*
-* Default OpenTelemetry histogram buckets target HTTP latency (0ms–10s)
-* and small counter values. LLM workloads have very different shapes:
-*
-*   - **Duration**: single-token prompts can be fast (50ms), long
-*     generations and reasoning models can run for minutes. Default buckets
-*     crush everything above 10s into one bucket.
-*   - **Token usage**: heavily right-skewed. A single request can range
-*     from tens of tokens to the million-token context windows.
-*   - **Cost (USD)**: per-request values are tiny (fractions of a cent),
-*     so linear buckets waste resolution at the low end.
-*
-* This module exposes empirically-chosen bucket arrays and a View helper
-* so users can apply them to their `MeterProvider` without knowing the
-* exact instrument names emitted by OpenAI/Anthropic/Traceloop plugins.
-*
-* @example
-* ```typescript
-* import { NodeSDK } from '@opentelemetry/sdk-node';
-* import { genAiMetricViews } from 'autotel';
-*
-* const sdk = new NodeSDK({
-*   serviceName: 'my-agent',
-*   views: [...genAiMetricViews()],
-* });
-* sdk.start();
-* ```
-*/
-/**
-* Duration buckets for LLM operations, in **seconds**. Covers fast
-* completions (50ms) through long-running reasoning jobs (5 min).
-*
-* Aligns with the OTel GenAI semantic conventions' published advice for
-* `gen_ai.client.operation.duration`.
-*/
-const GEN_AI_DURATION_BUCKETS_SECONDS = Object.freeze([
-	.01,
-	.05,
-	.1,
-	.25,
-	.5,
-	1,
-	2,
-	5,
-	10,
-	20,
-	30,
-	60,
-	120,
-	300
-]);
-/**
-* Token-count buckets for prompt, completion, and total token histograms.
-* Ranges from tiny prompts to million-token context windows.
-*
-* Aligns with the OTel GenAI semantic conventions' published advice for
-* `gen_ai.client.token.usage`.
-*/
-const GEN_AI_TOKEN_USAGE_BUCKETS = Object.freeze([
-	1,
-	4,
-	16,
-	64,
-	256,
-	1024,
-	4096,
-	16384,
-	65536,
-	262144,
-	1048576,
-	4194304
-]);
-/**
-* USD cost buckets. Sub-cent resolution at the low end (fractions of a
-* cent per small call) up to tens of dollars (batch jobs, Opus/o1 runs).
-*/
-const GEN_AI_COST_USD_BUCKETS = Object.freeze([
-	1e-5,
-	1e-4,
-	.001,
-	.005,
-	.01,
-	.05,
-	.1,
-	.5,
-	1,
-	5,
-	10,
-	50
-]);
-/**
-* Instrument-level advice object for `createHistogram(name, advice)`.
-* Use when you control the instrument creation (e.g. custom business
-* LLM metrics); `genAiMetricViews()` is better when the metric comes
-* from a third-party plugin.
-*/
-function llmHistogramAdvice(kind) {
-	return { advice: { explicitBucketBoundaries: [...kind === "duration" ? GEN_AI_DURATION_BUCKETS_SECONDS : kind === "tokens" ? GEN_AI_TOKEN_USAGE_BUCKETS : GEN_AI_COST_USD_BUCKETS] } };
-}
-/**
-* Returns `View`s that re-bucket the standard OTel GenAI histograms. Pass
-* the result to your `MeterProvider`'s `views` option.
-*
-* Matches instrument names emitted by:
-* - OpenTelemetry GenAI autoinstrumentation
-* - OpenInference / OpenLLMetry (traceloop)
-* - Arize Phoenix, LangSmith, etc. that follow the OTel spec
-*
-* Add more instrument patterns via the `extra` argument if you emit
-* custom LLM metrics.
-*/
-function genAiMetricViews(extra = []) {
-	return [...[
-		{
-			instrumentName: "gen_ai.client.operation.duration",
-			kind: "duration"
-		},
-		{
-			instrumentName: "gen_ai.client.token.usage",
-			kind: "tokens"
-		},
-		{
-			instrumentName: "gen_ai.client.cost.usd",
-			kind: "cost"
-		}
-	], ...extra].map(({ instrumentName, kind }) => ({
-		instrumentName,
-		aggregation: {
-			type: _opentelemetry_sdk_metrics.AggregationType.EXPLICIT_BUCKET_HISTOGRAM,
-			options: { boundaries: kind === "duration" ? [...GEN_AI_DURATION_BUCKETS_SECONDS] : kind === "tokens" ? [...GEN_AI_TOKEN_USAGE_BUCKETS] : [...GEN_AI_COST_USD_BUCKETS] }
-		}
-	}));
-}
-
-//#endregion
-//#region src/gen-ai-events.ts
-/**
-* Record that a prompt was dispatched to the provider. Typically called
-* before `await provider.chat.completions.create(...)`.
-*/
-function recordPromptSent(ctx, event = {}) {
-	require_correlated_events.emitCorrelatedEvent(ctx, "gen_ai.prompt.sent", buildPromptSentAttrs(event));
-}
-/**
-* Record a successful provider response. Call after the response arrives
-* (for non-streaming) or after the stream completes.
-*/
-function recordResponseReceived(ctx, event = {}) {
-	require_correlated_events.emitCorrelatedEvent(ctx, "gen_ai.response.received", buildResponseAttrs(event));
-}
-/**
-* Record a retry attempt on an LLM call. Call *before* sleeping for
-* `delayMs` so the event timestamp accurately marks when the retry
-* decision was made.
-*/
-function recordRetry(ctx, event) {
-	require_correlated_events.emitCorrelatedEvent(ctx, "gen_ai.retry", buildRetryAttrs(event));
-}
-/**
-* Record a tool / function call made in the course of an agent step.
-* Emits an event rather than a child span because many frameworks fire
-* several tool calls within a single provider response.
-*/
-function recordToolCall(ctx, event) {
-	require_correlated_events.emitCorrelatedEvent(ctx, "gen_ai.tool.call", buildToolCallAttrs(event));
-}
-/**
-* Record the time-to-first-token for a streaming response. Pair with
-* `recordResponseReceived` at the end so the span carries both the TTFT
-* marker and the final usage numbers.
-*/
-function recordStreamFirstToken(ctx, event = {}) {
-	require_correlated_events.emitCorrelatedEvent(ctx, "gen_ai.stream.first_token", buildStreamFirstTokenAttrs(event));
-}
-function buildPromptSentAttrs(event) {
-	const attrs = {};
-	if (event.model) attrs["gen_ai.request.model"] = event.model;
-	if (event.promptTokens !== void 0) attrs["gen_ai.usage.input_tokens"] = event.promptTokens;
-	if (event.messageCount !== void 0) attrs["gen_ai.request.message_count"] = event.messageCount;
-	if (event.operation) attrs["gen_ai.operation.name"] = event.operation;
-	return attrs;
-}
-function buildResponseAttrs(event) {
-	const attrs = {};
-	if (event.model) attrs["gen_ai.response.model"] = event.model;
-	if (event.promptTokens !== void 0) attrs["gen_ai.usage.input_tokens"] = event.promptTokens;
-	if (event.completionTokens !== void 0) attrs["gen_ai.usage.output_tokens"] = event.completionTokens;
-	if (event.totalTokens !== void 0) attrs["gen_ai.usage.total_tokens"] = event.totalTokens;
-	if (event.finishReasons && event.finishReasons.length > 0) attrs["gen_ai.response.finish_reasons"] = event.finishReasons.join(",");
-	return attrs;
-}
-function buildRetryAttrs(event) {
-	const attrs = { "retry.attempt": event.attempt };
-	if (event.reason) attrs["retry.reason"] = event.reason;
-	if (event.delayMs !== void 0) attrs["retry.delay_ms"] = event.delayMs;
-	if (event.statusCode !== void 0) attrs["http.response.status_code"] = event.statusCode;
-	return attrs;
-}
-function buildToolCallAttrs(event) {
-	const attrs = { "gen_ai.tool.name": event.toolName };
-	if (event.toolCallId) attrs["gen_ai.tool.call.id"] = event.toolCallId;
-	if (event.arguments) attrs["gen_ai.tool.arguments"] = event.arguments;
-	return attrs;
-}
-function buildStreamFirstTokenAttrs(event) {
-	const attrs = {};
-	if (event.tokensSoFar !== void 0) attrs["gen_ai.stream.tokens_so_far"] = event.tokensSoFar;
-	return attrs;
-}
-
-//#endregion
-//#region src/gen-ai-cost.ts
-/** Span attribute key autotel sets for an estimated call cost. */
-const GEN_AI_COST_ATTRIBUTE = "gen_ai.usage.cost.usd";
-/**
-* Approximate public list prices (USD per 1M tokens) at the time of writing.
-* Prices change; treat these as convenience defaults, not a billing source of
-* truth. Override per call via `options.pricing` or mutate this table at init.
-* Matching is exact first, then by longest key prefix, so versioned model ids
-* (`claude-sonnet-4-6-20251101`) resolve to a base entry (`claude-sonnet-4-6`).
-*/
-const MODEL_PRICING = {
-	"gpt-4o": {
-		inputPer1M: 2.5,
-		outputPer1M: 10
-	},
-	"gpt-4o-mini": {
-		inputPer1M: .15,
-		outputPer1M: .6
-	},
-	"gpt-4.1": {
-		inputPer1M: 2,
-		outputPer1M: 8
-	},
-	"gpt-4.1-mini": {
-		inputPer1M: .4,
-		outputPer1M: 1.6
-	},
-	"gpt-4.1-nano": {
-		inputPer1M: .1,
-		outputPer1M: .4
-	},
-	"o3-mini": {
-		inputPer1M: 1.1,
-		outputPer1M: 4.4
-	},
-	"claude-opus-4": {
-		inputPer1M: 15,
-		outputPer1M: 75
-	},
-	"claude-sonnet-4": {
-		inputPer1M: 3,
-		outputPer1M: 15
-	},
-	"claude-3-5-sonnet": {
-		inputPer1M: 3,
-		outputPer1M: 15
-	},
-	"claude-3-5-haiku": {
-		inputPer1M: .8,
-		outputPer1M: 4
-	},
-	"claude-3-opus": {
-		inputPer1M: 15,
-		outputPer1M: 75
-	},
-	"claude-3-haiku": {
-		inputPer1M: .25,
-		outputPer1M: 1.25
-	},
-	"gemini-1.5-pro": {
-		inputPer1M: 1.25,
-		outputPer1M: 5
-	},
-	"gemini-1.5-flash": {
-		inputPer1M: .075,
-		outputPer1M: .3
-	},
-	"gemini-2.0-flash": {
-		inputPer1M: .1,
-		outputPer1M: .4
-	}
-};
-function resolvePricing(table, model) {
-	const exact = table[model];
-	if (exact) return exact;
-	let best;
-	let bestLength = 0;
-	for (const key of Object.keys(table)) if (model.startsWith(key) && key.length > bestLength) {
-		best = table[key];
-		bestLength = key.length;
-	}
-	return best;
-}
-function round(value) {
-	return Math.round(value * 1e6) / 1e6;
-}
-/**
-* Estimate the USD cost of an LLM call. Returns `undefined` when the model has
-* no known pricing (supply one via `options.pricing`).
-*/
-function estimateLLMCost(model, usage, options) {
-	const price = resolvePricing(options?.pricing ? {
-		...MODEL_PRICING,
-		...options.pricing
-	} : MODEL_PRICING, model);
-	if (!price) return void 0;
-	const cachedInput = usage.cachedInputTokens ?? 0;
-	const billedInput = Math.max(0, (usage.inputTokens ?? 0) - cachedInput);
-	const output = usage.outputTokens ?? 0;
-	const cachedRate = price.cachedInputPer1M ?? price.inputPer1M;
-	return round(billedInput / 1e6 * price.inputPer1M + cachedInput / 1e6 * cachedRate + output / 1e6 * price.outputPer1M);
-}
-/**
-* Estimate cost and record it on `ctx` as the `gen_ai.usage.cost.usd` span
-* attribute. Returns the estimated cost, or `undefined` when the model is
-* unknown (in which case no attribute is set).
-*/
-function recordLLMCost(ctx, model, usage, options) {
-	const cost = estimateLLMCost(model, usage, options);
-	if (cost !== void 0) ctx.setAttribute(GEN_AI_COST_ATTRIBUTE, cost);
-	return cost;
-}
-
 //#endregion
 exports.AUTOTEL_SAMPLING_TAIL_EVALUATED = require_sampling.AUTOTEL_SAMPLING_TAIL_EVALUATED;
 exports.AUTOTEL_SAMPLING_TAIL_KEEP = require_sampling.AUTOTEL_SAMPLING_TAIL_KEEP;
@@ -982,12 +653,7 @@ exports.BusinessBaggage = require_business_baggage.BusinessBaggage;
 exports.CORRELATION_ID_BAGGAGE_KEY = require_track.CORRELATION_ID_BAGGAGE_KEY;
 exports.Event = require_event.Event;
 exports.FilteringSpanProcessor = require_filtering_span_processor.FilteringSpanProcessor;
-exports.GEN_AI_COST_ATTRIBUTE = GEN_AI_COST_ATTRIBUTE;
-exports.GEN_AI_COST_USD_BUCKETS = GEN_AI_COST_USD_BUCKETS;
-exports.GEN_AI_DURATION_BUCKETS_SECONDS = GEN_AI_DURATION_BUCKETS_SECONDS;
-exports.GEN_AI_TOKEN_USAGE_BUCKETS = GEN_AI_TOKEN_USAGE_BUCKETS;
 exports.HTTPAttributes = require_registry.HTTPAttributes;
-exports.MODEL_PRICING = MODEL_PRICING;
 exports.Metric = require_metric.Metric;
 exports.NORMALIZER_PATTERNS = require_span_name_normalizer.NORMALIZER_PATTERNS;
 exports.NORMALIZER_PRESETS = require_span_name_normalizer.NORMALIZER_PRESETS;
@@ -1049,14 +715,12 @@ exports.defineErrorCatalog = defineErrorCatalog;
 exports.defineEvent = defineEvent;
 exports.defineHttpDrain = defineHttpDrain;
 exports.enrichWithTraceContext = require_trace_helpers.enrichWithTraceContext;
-exports.estimateLLMCost = estimateLLMCost;
 exports.extractLinksFromBatch = require_sampling.extractLinksFromBatch;
 exports.finalizeSpan = require_trace_helpers.finalizeSpan;
 exports.flattenMetadata = require_trace_helpers.flattenMetadata;
 exports.flattenToAttributes = require_structured_error.flattenToAttributes;
 exports.flush = flush;
 exports.formatDuration = require_canonical_log_line_processor.formatDuration;
-exports.genAiMetricViews = genAiMetricViews;
 exports.generateCorrelationId = require_track.generateCorrelationId;
 exports.getActiveContext = require_trace_helpers.getActiveContext;
 exports.getActiveSpan = require_trace_helpers.getActiveSpan;
@@ -1089,7 +753,6 @@ exports.isInWorkflow = require_workflow.isInWorkflow;
 exports.isInitialized = require_init.isInitialized;
 exports.isLoggerLocked = require_init.isLoggerLocked;
 exports.isTracing = require_trace_helpers.isTracing;
-exports.llmHistogramAdvice = llmHistogramAdvice;
 exports.lockLogger = require_init.lockLogger;
 exports.markAsImmediate = require_functional.markAsImmediate;
 exports.mergeAttrs = require_attributes.mergeAttrs;
@@ -1108,13 +771,7 @@ Object.defineProperty(exports, 'propagation', {
     return _opentelemetry_api.propagation;
   }
 });
-exports.recordLLMCost = recordLLMCost;
-exports.recordPromptSent = recordPromptSent;
-exports.recordResponseReceived = recordResponseReceived;
-exports.recordRetry = recordRetry;
-exports.recordStreamFirstToken = recordStreamFirstToken;
 exports.recordStructuredError = require_structured_error.recordStructuredError;
-exports.recordToolCall = recordToolCall;
 exports.request = require_attributes.request;
 exports.resetEvents = require_event.resetEvents;
 exports.resetMetrics = require_metric.resetMetrics;
@@ -1142,7 +799,6 @@ exports.trace = trace;
 exports.traceConsumer = require_messaging.traceConsumer;
 exports.traceDB = require_semantic_helpers.traceDB;
 exports.traceHTTP = require_semantic_helpers.traceHTTP;
-exports.traceLLM = require_semantic_helpers.traceLLM;
 exports.traceMessaging = require_semantic_helpers.traceMessaging;
 exports.traceProducer = require_messaging.traceProducer;
 exports.traceStep = require_workflow.traceStep;