@drawbridge/drawbridge-utils 0.0.33 → 0.0.36

package/dist/ai.cjs

@@ -24,10 +24,13 @@ __export(ai_exports, {
   costForRequest: () => costForRequest,
   google: () => google,
   priceForRequest: () => priceForRequest,
-  pricing: () => pricing
+  pricing: () => pricing,
+  toolCost: () => toolCost,
+  toolPricing: () => toolPricing
 });
 module.exports = __toCommonJS(ai_exports);
 var import_genai = require("@google/genai");
+var import_drawbridge_telemetry = require("@drawbridge/drawbridge-telemetry");
 
 // circuit.js
 var CLOSED = "CLOSED";
@@ -176,6 +179,15 @@ var cost = {
   }
 };
 var MARKUP = 1.3;
+var toolCost = {
+  search: 3.5
+};
+var toolPricing = Object.fromEntries(
+  Object.entries(toolCost).map(([tool, value]) => [
+    tool,
+    Math.ceil(value * MARKUP)
+  ])
+);
 var pricing = Object.fromEntries(
   Object.entries(cost).map(([model, rates]) => [
     model,
@@ -187,35 +199,48 @@ var pricing = Object.fromEntries(
   ])
 );
 var sumCents = (rates, tokens) => Math.ceil(
-  (Number((tokens == null ? void 0 : tokens.cached) || 0) * rates.cached + Number((tokens == null ? void 0 : tokens.input) || 0) * rates.input + Number((tokens == null ? void 0 : tokens.output) || 0) * rates.output) / 1e6
+  (Number((tokens == null ? void 0 : tokens.cached) || 0) * rates.cached + Number((tokens == null ? void 0 : tokens.input) || 0) * rates.input + (Number((tokens == null ? void 0 : tokens.output) || 0) + Number((tokens == null ? void 0 : tokens.thinking) || 0)) * rates.output) / 1e6
 );
-var priceForRequest = (model, usage) => {
+var sumToolCents = (table, tools) => Object.entries(tools || {}).reduce(
+  (total, [name, used]) => total + (used && table[name] ? table[name] : 0),
+  0
+);
+var priceForRequest = (model, usage, tools) => {
   const rates = pricing[model];
   if (!rates) {
     throw new Error(`Unknown model for pricing: ${model}`);
   }
-  return sumCents(rates, usage);
+  return sumCents(rates, usage) + Math.ceil(sumToolCents(toolPricing, tools));
 };
-var costForRequest = (model, usage) => {
+var costForRequest = (model, usage, tools) => {
   const rates = cost[model];
   if (!rates) {
     throw new Error(`Unknown model for cost: ${model}`);
   }
-  return sumCents(rates, usage);
+  return sumCents(rates, usage) + Math.ceil(sumToolCents(toolCost, tools));
 };
 var tokensFromMetadata = (metadata) => ({
   cached: Number(metadata == null ? void 0 : metadata.cachedContentTokenCount) || 0,
   input: Number(metadata == null ? void 0 : metadata.promptTokenCount) || 0,
-  output: Number(metadata == null ? void 0 : metadata.candidatesTokenCount) || 0
+  output: Number(metadata == null ? void 0 : metadata.candidatesTokenCount) || 0,
+  thinking: Number(metadata == null ? void 0 : metadata.thoughtsTokenCount) || 0
 });
+var toolsFromResponse = (response) => {
+  var _a, _b;
+  return {
+    search: Boolean((_b = (_a = response == null ? void 0 : response.candidates) == null ? void 0 : _a[0]) == null ? void 0 : _b.groundingMetadata)
+  };
+};
 var billRequest = async ({
   db,
   user,
   model,
-  usage
+  usage,
+  tools
 }) => {
-  const gross = priceForRequest(model, usage);
-  const cogs = costForRequest(model, usage);
+  const gross = priceForRequest(model, usage, tools);
+  const cogs = costForRequest(model, usage, tools);
+  const id = (0, import_drawbridge_telemetry.currentTraceId)();
   await debit({
     db,
     user,
@@ -223,10 +248,12 @@ var billRequest = async ({
     type: "ai",
     category: "usage",
     source: "user",
+    ...id && { trace: { id } },
     ai: {
       name: "google",
       model,
       tokens: usage,
+      tools,
       rates: {
         wholesale: cost[model],
         retail: pricing[model]
@@ -271,7 +298,8 @@ var google = {
         db,
         user,
         model,
-        usage: tokensFromMetadata(response == null ? void 0 : response.usageMetadata)
+        usage: tokensFromMetadata(response == null ? void 0 : response.usageMetadata),
+        tools: toolsFromResponse(response)
       });
     }
     return { content };
@@ -301,7 +329,8 @@ var google = {
         db,
         user,
         model,
-        usage: tokensFromMetadata(response == null ? void 0 : response.usageMetadata)
+        usage: tokensFromMetadata(response == null ? void 0 : response.usageMetadata),
+        tools: toolsFromResponse(response)
       });
     }
     return { content };
@@ -314,5 +343,7 @@ var google = {
   costForRequest,
   google,
   priceForRequest,
-  pricing
+  pricing,
+  toolCost,
+  toolPricing
 });

package/dist/ai.d.cts

@@ -1,4 +1,5 @@
 import { GoogleGenAI } from '@google/genai';
+import { currentTraceId } from '@drawbridge/drawbridge-telemetry';
 import { circuit } from './circuit.cjs';
 import { debit } from './transactions.cjs';
 
@@ -61,6 +62,22 @@ const cost = {
 // refactor later if tier-based pricing becomes a real need.
 const MARKUP = 1.3;
 
+// Flat per-request fees Google charges on top of tokens when specific tools
+// fire. Search grounding is $35/1k requests → 3.5 cents/request. Billed only
+// when the response actually used the tool (response.candidates[0].
+// groundingMetadata is populated) — enabling the tool in config without it
+// being invoked is free. Add more tools here as Google introduces them.
+const toolCost = {
+	search : 3.5
+};
+
+const toolPricing = Object.fromEntries(
+	Object.entries( toolCost ).map( ( [ tool, value ] ) => [
+		tool,
+		Math.ceil( value * MARKUP )
+	] )
+);
+
 // Derived retail pricing in cents per 1,000,000 tokens — what users pay.
 // Rounded so the per-1M rate is a whole-cent value; per-request rounding
 // happens in priceForRequest via Math.ceil so we never under-bill.
@@ -75,17 +92,33 @@ const pricing = Object.fromEntries(
 	] )
 );
 
+// Gemini 2.5 emits internal reasoning ("thinking") tokens alongside the
+// visible response. Google bills them at the standard output rate, so they
+// roll into the same multiplier here even though they live as a distinct
+// field on the tokens object for visibility / margin analysis.
 const sumCents = ( rates, tokens ) => Math.ceil(
 	(
 		Number( tokens?.cached || 0 ) * rates.cached +
 		Number( tokens?.input || 0 ) * rates.input +
-		Number( tokens?.output || 0 ) * rates.output
+		(
+			Number( tokens?.output || 0 ) +
+			Number( tokens?.thinking || 0 )
+		) * rates.output
 	) / 1_000_000
 );
 
+// Sum the per-request flat fees for the tools that actually fired. `tools`
+// is the same { search: true } shape stamped on the transaction row, so the
+// row's `tools` object and the amount it contributes are recomputable from
+// the same source of truth.
+const sumToolCents = ( table, tools ) => Object.entries( tools || {} ).reduce(
+	( total, [ name, used ] ) => total + ( used && table[ name ] ? table[ name ] : 0 ),
+	0
+);
+
 // What we charge the user, rounded up so we never under-bill. Throws on an
 // unknown model so a typo doesn't silently fall back to a wrong price.
-const priceForRequest = ( model, usage ) => {
+const priceForRequest = ( model, usage, tools ) => {
 
 	const rates = pricing[ model ];
 
@@ -95,13 +128,13 @@ const priceForRequest = ( model, usage ) => {
 
 	}
 
-	return sumCents( rates, usage );
+	return sumCents( rates, usage ) + Math.ceil( sumToolCents( toolPricing, tools ) );
 
 };
 
 // What we pay the provider. Recorded on each transaction row so margin is
 // computable per request without backfilling from git history.
-const costForRequest = ( model, usage ) => {
+const costForRequest = ( model, usage, tools ) => {
 
 	const rates = cost[ model ];
 
@@ -111,15 +144,27 @@ const costForRequest = ( model, usage ) => {
 
 	}
 
-	return sumCents( rates, usage );
+	return sumCents( rates, usage ) + Math.ceil( sumToolCents( toolCost, tools ) );
 
 };
 
 // Map Google's usageMetadata shape into the pricing.tokens shape.
+// `thinking` is Gemini 2.5's internal reasoning token count — billed at the
+// same rate as `output` but tracked separately so the transaction row shows
+// the visible-vs-reasoning split.
 const tokensFromMetadata = ( metadata ) => ({
 	cached : Number( metadata?.cachedContentTokenCount ) || 0,
 	input : Number( metadata?.promptTokenCount ) || 0,
-	output : Number( metadata?.candidatesTokenCount ) || 0
+	output : Number( metadata?.candidatesTokenCount ) || 0,
+	thinking : Number( metadata?.thoughtsTokenCount ) || 0
+});
+
+// Derive which billable tools actually fired from the response. Google bills
+// per-request when the tool was invoked, not when it was merely enabled in
+// config — `groundingMetadata` is only present on candidates when search was
+// used. Add more tools here as Google adds them.
+const toolsFromResponse = ( response ) => ({
+	search : Boolean( response?.candidates?.[ 0 ]?.groundingMetadata )
 });
 
 // Compute gross + cost from response usage and atomically debit the user.
@@ -131,11 +176,22 @@ const billRequest = async ({
 	db,
 	user,
 	model,
-	usage
+	usage,
+	tools
 }) => {
 
-	const gross = priceForRequest( model, usage );
-	const cogs = costForRequest( model, usage );
+	const gross = priceForRequest( model, usage, tools );
+	const cogs = costForRequest( model, usage, tools );
+
+	// Stamp the current trace ID so multiple AI calls inside one HTTP request
+	// (or one BullMQ job) share a grouping key — the UI can collapse rows
+	// per trace.id to show "this preview cost $0.12 across 5 calls" without
+	// the user needing to mentally fold them together. currentTraceId()
+	// returns undefined when called outside any traced scope, in which case
+	// the field is omitted from the row (additionalProperties:false would
+	// reject `trace: undefined`). Nested as `trace.id` to match the `ai` /
+	// `balance` sub-object pattern already on this document.
+	const id = currentTraceId();
 
 	await debit({
 		db,
@@ -144,10 +200,12 @@ const billRequest = async ({
 		type : 'ai',
 		category : 'usage',
 		source : 'user',
+		...( id && { trace : { id } }),
 		ai : {
 			name : 'google',
 			model,
 			tokens : usage,
+			tools,
 			rates : {
 				wholesale : cost[ model ],
 				retail : pricing[ model ]
@@ -206,7 +264,8 @@ const google = {
 				db,
 				user,
 				model,
-				usage : tokensFromMetadata( response?.usageMetadata )
+				usage : tokensFromMetadata( response?.usageMetadata ),
+				tools : toolsFromResponse( response )
 			});
 
 		}
@@ -251,7 +310,8 @@ const google = {
 				db,
 				user,
 				model,
-				usage : tokensFromMetadata( response?.usageMetadata )
+				usage : tokensFromMetadata( response?.usageMetadata ),
+				tools : toolsFromResponse( response )
 			});
 
 		}
@@ -262,4 +322,4 @@ const google = {
 
 };
 
-export { MARKUP, cost, costForRequest, google, priceForRequest, pricing };
+export { MARKUP, cost, costForRequest, google, priceForRequest, pricing, toolCost, toolPricing };

package/dist/ai.d.ts

@@ -1,4 +1,5 @@
 import { GoogleGenAI } from '@google/genai';
+import { currentTraceId } from '@drawbridge/drawbridge-telemetry';
 import { circuit } from './circuit.js';
 import { debit } from './transactions.js';
 
@@ -61,6 +62,22 @@ const cost = {
 // refactor later if tier-based pricing becomes a real need.
 const MARKUP = 1.3;
 
+// Flat per-request fees Google charges on top of tokens when specific tools
+// fire. Search grounding is $35/1k requests → 3.5 cents/request. Billed only
+// when the response actually used the tool (response.candidates[0].
+// groundingMetadata is populated) — enabling the tool in config without it
+// being invoked is free. Add more tools here as Google introduces them.
+const toolCost = {
+	search : 3.5
+};
+
+const toolPricing = Object.fromEntries(
+	Object.entries( toolCost ).map( ( [ tool, value ] ) => [
+		tool,
+		Math.ceil( value * MARKUP )
+	] )
+);
+
 // Derived retail pricing in cents per 1,000,000 tokens — what users pay.
 // Rounded so the per-1M rate is a whole-cent value; per-request rounding
 // happens in priceForRequest via Math.ceil so we never under-bill.
@@ -75,17 +92,33 @@ const pricing = Object.fromEntries(
 	] )
 );
 
+// Gemini 2.5 emits internal reasoning ("thinking") tokens alongside the
+// visible response. Google bills them at the standard output rate, so they
+// roll into the same multiplier here even though they live as a distinct
+// field on the tokens object for visibility / margin analysis.
 const sumCents = ( rates, tokens ) => Math.ceil(
 	(
 		Number( tokens?.cached || 0 ) * rates.cached +
 		Number( tokens?.input || 0 ) * rates.input +
-		Number( tokens?.output || 0 ) * rates.output
+		(
+			Number( tokens?.output || 0 ) +
+			Number( tokens?.thinking || 0 )
+		) * rates.output
 	) / 1_000_000
 );
 
+// Sum the per-request flat fees for the tools that actually fired. `tools`
+// is the same { search: true } shape stamped on the transaction row, so the
+// row's `tools` object and the amount it contributes are recomputable from
+// the same source of truth.
+const sumToolCents = ( table, tools ) => Object.entries( tools || {} ).reduce(
+	( total, [ name, used ] ) => total + ( used && table[ name ] ? table[ name ] : 0 ),
+	0
+);
+
 // What we charge the user, rounded up so we never under-bill. Throws on an
 // unknown model so a typo doesn't silently fall back to a wrong price.
-const priceForRequest = ( model, usage ) => {
+const priceForRequest = ( model, usage, tools ) => {
 
 	const rates = pricing[ model ];
 
@@ -95,13 +128,13 @@ const priceForRequest = ( model, usage ) => {
 
 	}
 
-	return sumCents( rates, usage );
+	return sumCents( rates, usage ) + Math.ceil( sumToolCents( toolPricing, tools ) );
 
 };
 
 // What we pay the provider. Recorded on each transaction row so margin is
 // computable per request without backfilling from git history.
-const costForRequest = ( model, usage ) => {
+const costForRequest = ( model, usage, tools ) => {
 
 	const rates = cost[ model ];
 
@@ -111,15 +144,27 @@ const costForRequest = ( model, usage ) => {
 
 	}
 
-	return sumCents( rates, usage );
+	return sumCents( rates, usage ) + Math.ceil( sumToolCents( toolCost, tools ) );
 
 };
 
 // Map Google's usageMetadata shape into the pricing.tokens shape.
+// `thinking` is Gemini 2.5's internal reasoning token count — billed at the
+// same rate as `output` but tracked separately so the transaction row shows
+// the visible-vs-reasoning split.
 const tokensFromMetadata = ( metadata ) => ({
 	cached : Number( metadata?.cachedContentTokenCount ) || 0,
 	input : Number( metadata?.promptTokenCount ) || 0,
-	output : Number( metadata?.candidatesTokenCount ) || 0
+	output : Number( metadata?.candidatesTokenCount ) || 0,
+	thinking : Number( metadata?.thoughtsTokenCount ) || 0
+});
+
+// Derive which billable tools actually fired from the response. Google bills
+// per-request when the tool was invoked, not when it was merely enabled in
+// config — `groundingMetadata` is only present on candidates when search was
+// used. Add more tools here as Google adds them.
+const toolsFromResponse = ( response ) => ({
+	search : Boolean( response?.candidates?.[ 0 ]?.groundingMetadata )
 });
 
 // Compute gross + cost from response usage and atomically debit the user.
@@ -131,11 +176,22 @@ const billRequest = async ({
 	db,
 	user,
 	model,
-	usage
+	usage,
+	tools
 }) => {
 
-	const gross = priceForRequest( model, usage );
-	const cogs = costForRequest( model, usage );
+	const gross = priceForRequest( model, usage, tools );
+	const cogs = costForRequest( model, usage, tools );
+
+	// Stamp the current trace ID so multiple AI calls inside one HTTP request
+	// (or one BullMQ job) share a grouping key — the UI can collapse rows
+	// per trace.id to show "this preview cost $0.12 across 5 calls" without
+	// the user needing to mentally fold them together. currentTraceId()
+	// returns undefined when called outside any traced scope, in which case
+	// the field is omitted from the row (additionalProperties:false would
+	// reject `trace: undefined`). Nested as `trace.id` to match the `ai` /
+	// `balance` sub-object pattern already on this document.
+	const id = currentTraceId();
 
 	await debit({
 		db,
@@ -144,10 +200,12 @@ const billRequest = async ({
 		type : 'ai',
 		category : 'usage',
 		source : 'user',
+		...( id && { trace : { id } }),
 		ai : {
 			name : 'google',
 			model,
 			tokens : usage,
+			tools,
 			rates : {
 				wholesale : cost[ model ],
 				retail : pricing[ model ]
@@ -206,7 +264,8 @@ const google = {
 				db,
 				user,
 				model,
-				usage : tokensFromMetadata( response?.usageMetadata )
+				usage : tokensFromMetadata( response?.usageMetadata ),
+				tools : toolsFromResponse( response )
 			});
 
 		}
@@ -251,7 +310,8 @@ const google = {
 				db,
 				user,
 				model,
-				usage : tokensFromMetadata( response?.usageMetadata )
+				usage : tokensFromMetadata( response?.usageMetadata ),
+				tools : toolsFromResponse( response )
 			});
 
 		}
@@ -262,4 +322,4 @@ const google = {
 
 };
 
-export { MARKUP, cost, costForRequest, google, priceForRequest, pricing };
+export { MARKUP, cost, costForRequest, google, priceForRequest, pricing, toolCost, toolPricing };

package/dist/ai.js

@@ -7,6 +7,7 @@ import {
 
 // ai.js
 import { GoogleGenAI } from "@google/genai";
+import { currentTraceId } from "@drawbridge/drawbridge-telemetry";
 var google_client = new GoogleGenAI({
   apiKey: process.env.GOOGLE_API_KEY
 });
@@ -32,6 +33,15 @@ var cost = {
   }
 };
 var MARKUP = 1.3;
+var toolCost = {
+  search: 3.5
+};
+var toolPricing = Object.fromEntries(
+  Object.entries(toolCost).map(([tool, value]) => [
+    tool,
+    Math.ceil(value * MARKUP)
+  ])
+);
 var pricing = Object.fromEntries(
   Object.entries(cost).map(([model, rates]) => [
     model,
@@ -43,35 +53,48 @@ var pricing = Object.fromEntries(
   ])
 );
 var sumCents = (rates, tokens) => Math.ceil(
-  (Number((tokens == null ? void 0 : tokens.cached) || 0) * rates.cached + Number((tokens == null ? void 0 : tokens.input) || 0) * rates.input + Number((tokens == null ? void 0 : tokens.output) || 0) * rates.output) / 1e6
+  (Number((tokens == null ? void 0 : tokens.cached) || 0) * rates.cached + Number((tokens == null ? void 0 : tokens.input) || 0) * rates.input + (Number((tokens == null ? void 0 : tokens.output) || 0) + Number((tokens == null ? void 0 : tokens.thinking) || 0)) * rates.output) / 1e6
 );
-var priceForRequest = (model, usage) => {
+var sumToolCents = (table, tools) => Object.entries(tools || {}).reduce(
+  (total, [name, used]) => total + (used && table[name] ? table[name] : 0),
+  0
+);
+var priceForRequest = (model, usage, tools) => {
   const rates = pricing[model];
   if (!rates) {
     throw new Error(`Unknown model for pricing: ${model}`);
   }
-  return sumCents(rates, usage);
+  return sumCents(rates, usage) + Math.ceil(sumToolCents(toolPricing, tools));
 };
-var costForRequest = (model, usage) => {
+var costForRequest = (model, usage, tools) => {
   const rates = cost[model];
   if (!rates) {
     throw new Error(`Unknown model for cost: ${model}`);
   }
-  return sumCents(rates, usage);
+  return sumCents(rates, usage) + Math.ceil(sumToolCents(toolCost, tools));
 };
 var tokensFromMetadata = (metadata) => ({
   cached: Number(metadata == null ? void 0 : metadata.cachedContentTokenCount) || 0,
   input: Number(metadata == null ? void 0 : metadata.promptTokenCount) || 0,
-  output: Number(metadata == null ? void 0 : metadata.candidatesTokenCount) || 0
+  output: Number(metadata == null ? void 0 : metadata.candidatesTokenCount) || 0,
+  thinking: Number(metadata == null ? void 0 : metadata.thoughtsTokenCount) || 0
 });
+var toolsFromResponse = (response) => {
+  var _a, _b;
+  return {
+    search: Boolean((_b = (_a = response == null ? void 0 : response.candidates) == null ? void 0 : _a[0]) == null ? void 0 : _b.groundingMetadata)
+  };
+};
 var billRequest = async ({
   db,
   user,
   model,
-  usage
+  usage,
+  tools
 }) => {
-  const gross = priceForRequest(model, usage);
-  const cogs = costForRequest(model, usage);
+  const gross = priceForRequest(model, usage, tools);
+  const cogs = costForRequest(model, usage, tools);
+  const id = currentTraceId();
   await debit({
     db,
     user,
@@ -79,10 +102,12 @@ var billRequest = async ({
     type: "ai",
     category: "usage",
     source: "user",
+    ...id && { trace: { id } },
     ai: {
       name: "google",
       model,
       tokens: usage,
+      tools,
       rates: {
         wholesale: cost[model],
         retail: pricing[model]
@@ -127,7 +152,8 @@ var google = {
         db,
         user,
         model,
-        usage: tokensFromMetadata(response == null ? void 0 : response.usageMetadata)
+        usage: tokensFromMetadata(response == null ? void 0 : response.usageMetadata),
+        tools: toolsFromResponse(response)
       });
     }
     return { content };
@@ -157,7 +183,8 @@ var google = {
         db,
         user,
         model,
-        usage: tokensFromMetadata(response == null ? void 0 : response.usageMetadata)
+        usage: tokensFromMetadata(response == null ? void 0 : response.usageMetadata),
+        tools: toolsFromResponse(response)
       });
     }
     return { content };
@@ -169,5 +196,7 @@ export {
   costForRequest,
   google,
   priceForRequest,
-  pricing
+  pricing,
+  toolCost,
+  toolPricing
 };

package/package.json

@@ -2,6 +2,7 @@
   "type": "module",
   "dependencies": {
     "@drawbridge/drawbridge-agents": "0.0.10",
+    "@drawbridge/drawbridge-telemetry": "0.0.13",
     "@google/genai": "1.30.0",
     "axios": "1.16.0",
     "currency-codes": "2.2.0",
@@ -103,5 +104,5 @@
     "build": "tsup && npm publish"
   },
   "types": "dist/index.d.ts",
-  "version": "0.0.33"
+  "version": "0.0.36"
 }