@drawbridge/drawbridge-utils 0.0.64 → 0.0.66

package/dist/ai.cjs

@@ -20,13 +20,7 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 var ai_exports = {};
 __export(ai_exports, {
   MARKUP: () => MARKUP,
-  cost: () => cost,
-  costForRequest: () => costForRequest,
-  google: () => google,
-  priceForRequest: () => priceForRequest,
-  pricing: () => pricing,
-  toolCost: () => toolCost,
-  toolPricing: () => toolPricing
+  google: () => google
 });
 module.exports = __toCommonJS(ai_exports);
 var import_genai = require("@google/genai");
@@ -155,19 +149,10 @@ var debit = async ({
   });
 };
 
-// lib/ai.js
-var google_client = new import_genai.GoogleGenAI({
-  apiKey: process.env.GOOGLE_GEMINI_API_KEY
-});
-var googleCircuit = circuit({
-  name: "google-ai",
-  threshold: 3,
-  timeout: 6e4
-});
-var models = {
-  image: "gemini-2.5-flash-image",
-  text: "gemini-2.5-flash"
-};
+// lib/billing.js
+var import_drawbridge_telemetry2 = require("@drawbridge/drawbridge-telemetry");
+var logger = (0, import_drawbridge_telemetry2.createLogger)();
+var MARKUP = 1.3;
 var cost = {
   "gemini-2.5-flash": {
     cached: 3,
@@ -180,7 +165,6 @@ var cost = {
     output: 3e3
   }
 };
-var MARKUP = 1.3;
 var toolCost = {
   search: 3.5
 };
@@ -221,25 +205,7 @@ var costForRequest = (model, usage, tools) => {
   }
   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,
-  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,
-  tools
-}) => {
+var billRequest = async ({ db, user, model, usage, tools }) => {
   const gross = priceForRequest(model, usage, tools);
   const cogs = costForRequest(model, usage, tools);
   await debit({
@@ -266,6 +232,42 @@ var billRequest = async ({
     }
   });
 };
+var ai = {
+  // Retail cents for a request (what the user pays).
+  price: priceForRequest,
+  // Wholesale cents for a request (what we pay the provider).
+  cost: costForRequest,
+  // Deferred bill() the caller fires on success, so the debit lands only when
+  // the request succeeds and bundles into that request's Sentry trace. No user
+  // → a no-op.
+  bill: ({ db, user, model, usage, tools }) => (user == null ? void 0 : user.id) ? () => billRequest({ db, user, model, usage, tools }) : () => Promise.resolve()
+};
+
+// lib/ai.js
+var google_client = new import_genai.GoogleGenAI({
+  apiKey: process.env.GOOGLE_GEMINI_API_KEY
+});
+var googleCircuit = circuit({
+  name: "google-ai",
+  threshold: 3,
+  timeout: 6e4
+});
+var models = {
+  image: "gemini-2.5-flash-image",
+  text: "gemini-2.5-flash"
+};
+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,
+  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 google = {
   image: async ({
     config = {},
@@ -297,7 +299,7 @@ var google = {
     }
     const usage = tokensFromMetadata(response == null ? void 0 : response.usageMetadata);
     const tools = toolsFromResponse(response);
-    const bill = (user == null ? void 0 : user.id) ? () => billRequest({ db, user, model, usage, tools }) : () => Promise.resolve();
+    const bill = ai.bill({ db, user, model, usage, tools });
     return { content, bill };
   },
   text: async ({
@@ -325,18 +327,12 @@ var google = {
     }
     const usage = tokensFromMetadata(response == null ? void 0 : response.usageMetadata);
     const tools = toolsFromResponse(response);
-    const bill = (user == null ? void 0 : user.id) ? () => billRequest({ db, user, model, usage, tools }) : () => Promise.resolve();
+    const bill = ai.bill({ db, user, model, usage, tools });
     return { content, bill };
   }
 };
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   MARKUP,
-  cost,
-  costForRequest,
-  google,
-  priceForRequest,
-  pricing,
-  toolCost,
-  toolPricing
+  google
 });

package/dist/ai.d.cts

@@ -1,7 +1,10 @@
 import { GoogleGenAI } from '@google/genai';
 import { circuit } from './circuit.cjs';
-import { debit } from './transactions.cjs';
+import { ai } from './billing.cjs';
+export { MARKUP } from './billing.cjs';
+import './transactions.cjs';
 import '@drawbridge/drawbridge-telemetry';
+import './usage.cjs';
 
 // AI request entry point. Wraps the Google GenAI client with a per-process
 // circuit breaker, owns the per-model pricing table, and atomically debits
@@ -37,117 +40,6 @@ const models = {
 	text : 'gemini-2.5-flash'
 };
 
-// Raw per-model provider rates in cents per 1,000,000 tokens. This is what
-// Drawbridge pays the provider — the wholesale cost. Sync from LiteLLM:
-// https://github.com/BerriAI/litellm/blob/main/litellm/model_prices_and_context_window_backup.json
-// Never apply markup here; the MARKUP constant below derives the retail
-// `pricing` table.
-const cost = {
-	'gemini-2.5-flash' : {
-		cached : 3,
-		input : 30,
-		output : 250
-	},
-	'gemini-2.5-flash-image' : {
-		cached : 3,
-		input : 30,
-		output : 3000
-	}
-};
-
-// Markup multiplier applied to every model's cost to produce the retail
-// price. 1.3 = 30% margin. Change one number, all pricing updates. Mirrors
-// the per-customer markup Stripe rate cards used to provide; keeping it
-// global (rather than per-model or per-plan) for simplicity — easy to
-// 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.
-const pricing = Object.fromEntries(
-	Object.entries( cost ).map( ( [ model, rates ] ) => [
-		model,
-		{
-			cached : Math.round( rates.cached * MARKUP ),
-			input : Math.round( rates.input * MARKUP ),
-			output : Math.round( rates.output * MARKUP )
-		}
-	] )
-);
-
-// 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 ) +
-			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, tools ) => {
-
-	const rates = pricing[ model ];
-
-	if( !rates ){
-
-		throw new Error( `Unknown model for pricing: ${ model }` );
-
-	}
-
-	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, tools ) => {
-
-	const rates = cost[ model ];
-
-	if( !rates ){
-
-		throw new Error( `Unknown model for cost: ${ model }` );
-
-	}
-
-	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
@@ -167,48 +59,6 @@ const toolsFromResponse = ( response ) => ({
 	search : Boolean( response?.candidates?.[ 0 ]?.groundingMetadata )
 });
 
-// Compute gross + cost from response usage and atomically debit the user.
-// Records the full income-statement line on the transaction row's `ai`
-// sub-doc: wholesale + retail per-1M rate snapshots, plus rolled-up cost,
-// net, and gross totals. Margin per request is `ai.net`; margin per query
-// is a single $sum: '$ai.net'.
-const billRequest = async ({
-	db,
-	user,
-	model,
-	usage,
-	tools
-}) => {
-
-	const gross = priceForRequest( model, usage, tools );
-	const cogs = costForRequest( model, usage, tools );
-
-	await debit({
-		db,
-		user,
-		amount : gross,
-		type : 'ai',
-		category : 'usage',
-		source : 'user',
-		ai : {
-			name : 'google',
-			model,
-			tokens : usage,
-			tools,
-			rates : {
-				wholesale : cost[ model ],
-				retail : pricing[ model ]
-			},
-			totals : {
-				cost : cogs,
-				net : gross - cogs,
-				gross
-			}
-		}
-	});
-
-};
-
 const google = {
 
 	image : async (
@@ -254,9 +104,7 @@ const google = {
 		const usage = tokensFromMetadata( response?.usageMetadata );
 		const tools = toolsFromResponse( response );
 
-		const bill = user?.id
-			? () => billRequest({ db, user, model, usage, tools })
-			: () => Promise.resolve();
+		const bill = ai.bill({ db, user, model, usage, tools });
 
 		return { content, bill };
 
@@ -310,9 +158,7 @@ const google = {
 		const usage = tokensFromMetadata( response?.usageMetadata );
 		const tools = toolsFromResponse( response );
 
-		const bill = user?.id
-			? () => billRequest({ db, user, model, usage, tools })
-			: () => Promise.resolve();
+		const bill = ai.bill({ db, user, model, usage, tools });
 
 		return { content, bill };
 
@@ -320,4 +166,4 @@ const google = {
 
 };
 
-export { MARKUP, cost, costForRequest, google, priceForRequest, pricing, toolCost, toolPricing };
+export { google };

package/dist/ai.d.ts

@@ -1,7 +1,10 @@
 import { GoogleGenAI } from '@google/genai';
 import { circuit } from './circuit.js';
-import { debit } from './transactions.js';
+import { ai } from './billing.js';
+export { MARKUP } from './billing.js';
+import './transactions.js';
 import '@drawbridge/drawbridge-telemetry';
+import './usage.js';
 
 // AI request entry point. Wraps the Google GenAI client with a per-process
 // circuit breaker, owns the per-model pricing table, and atomically debits
@@ -37,117 +40,6 @@ const models = {
 	text : 'gemini-2.5-flash'
 };
 
-// Raw per-model provider rates in cents per 1,000,000 tokens. This is what
-// Drawbridge pays the provider — the wholesale cost. Sync from LiteLLM:
-// https://github.com/BerriAI/litellm/blob/main/litellm/model_prices_and_context_window_backup.json
-// Never apply markup here; the MARKUP constant below derives the retail
-// `pricing` table.
-const cost = {
-	'gemini-2.5-flash' : {
-		cached : 3,
-		input : 30,
-		output : 250
-	},
-	'gemini-2.5-flash-image' : {
-		cached : 3,
-		input : 30,
-		output : 3000
-	}
-};
-
-// Markup multiplier applied to every model's cost to produce the retail
-// price. 1.3 = 30% margin. Change one number, all pricing updates. Mirrors
-// the per-customer markup Stripe rate cards used to provide; keeping it
-// global (rather than per-model or per-plan) for simplicity — easy to
-// 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.
-const pricing = Object.fromEntries(
-	Object.entries( cost ).map( ( [ model, rates ] ) => [
-		model,
-		{
-			cached : Math.round( rates.cached * MARKUP ),
-			input : Math.round( rates.input * MARKUP ),
-			output : Math.round( rates.output * MARKUP )
-		}
-	] )
-);
-
-// 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 ) +
-			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, tools ) => {
-
-	const rates = pricing[ model ];
-
-	if( !rates ){
-
-		throw new Error( `Unknown model for pricing: ${ model }` );
-
-	}
-
-	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, tools ) => {
-
-	const rates = cost[ model ];
-
-	if( !rates ){
-
-		throw new Error( `Unknown model for cost: ${ model }` );
-
-	}
-
-	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
@@ -167,48 +59,6 @@ const toolsFromResponse = ( response ) => ({
 	search : Boolean( response?.candidates?.[ 0 ]?.groundingMetadata )
 });
 
-// Compute gross + cost from response usage and atomically debit the user.
-// Records the full income-statement line on the transaction row's `ai`
-// sub-doc: wholesale + retail per-1M rate snapshots, plus rolled-up cost,
-// net, and gross totals. Margin per request is `ai.net`; margin per query
-// is a single $sum: '$ai.net'.
-const billRequest = async ({
-	db,
-	user,
-	model,
-	usage,
-	tools
-}) => {
-
-	const gross = priceForRequest( model, usage, tools );
-	const cogs = costForRequest( model, usage, tools );
-
-	await debit({
-		db,
-		user,
-		amount : gross,
-		type : 'ai',
-		category : 'usage',
-		source : 'user',
-		ai : {
-			name : 'google',
-			model,
-			tokens : usage,
-			tools,
-			rates : {
-				wholesale : cost[ model ],
-				retail : pricing[ model ]
-			},
-			totals : {
-				cost : cogs,
-				net : gross - cogs,
-				gross
-			}
-		}
-	});
-
-};
-
 const google = {
 
 	image : async (
@@ -254,9 +104,7 @@ const google = {
 		const usage = tokensFromMetadata( response?.usageMetadata );
 		const tools = toolsFromResponse( response );
 
-		const bill = user?.id
-			? () => billRequest({ db, user, model, usage, tools })
-			: () => Promise.resolve();
+		const bill = ai.bill({ db, user, model, usage, tools });
 
 		return { content, bill };
 
@@ -310,9 +158,7 @@ const google = {
 		const usage = tokensFromMetadata( response?.usageMetadata );
 		const tools = toolsFromResponse( response );
 
-		const bill = user?.id
-			? () => billRequest({ db, user, model, usage, tools })
-			: () => Promise.resolve();
+		const bill = ai.bill({ db, user, model, usage, tools });
 
 		return { content, bill };
 
@@ -320,4 +166,4 @@ const google = {
 
 };
 
-export { MARKUP, cost, costForRequest, google, priceForRequest, pricing, toolCost, toolPricing };
+export { google };

package/dist/ai.js

@@ -125,19 +125,10 @@ var debit = async ({
   });
 };
 
-// lib/ai.js
-var google_client = new GoogleGenAI({
-  apiKey: process.env.GOOGLE_GEMINI_API_KEY
-});
-var googleCircuit = circuit({
-  name: "google-ai",
-  threshold: 3,
-  timeout: 6e4
-});
-var models = {
-  image: "gemini-2.5-flash-image",
-  text: "gemini-2.5-flash"
-};
+// lib/billing.js
+import { createLogger } from "@drawbridge/drawbridge-telemetry";
+var logger = createLogger();
+var MARKUP = 1.3;
 var cost = {
   "gemini-2.5-flash": {
     cached: 3,
@@ -150,7 +141,6 @@ var cost = {
     output: 3e3
   }
 };
-var MARKUP = 1.3;
 var toolCost = {
   search: 3.5
 };
@@ -191,25 +181,7 @@ var costForRequest = (model, usage, tools) => {
   }
   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,
-  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,
-  tools
-}) => {
+var billRequest = async ({ db, user, model, usage, tools }) => {
   const gross = priceForRequest(model, usage, tools);
   const cogs = costForRequest(model, usage, tools);
   await debit({
@@ -236,6 +208,42 @@ var billRequest = async ({
     }
   });
 };
+var ai = {
+  // Retail cents for a request (what the user pays).
+  price: priceForRequest,
+  // Wholesale cents for a request (what we pay the provider).
+  cost: costForRequest,
+  // Deferred bill() the caller fires on success, so the debit lands only when
+  // the request succeeds and bundles into that request's Sentry trace. No user
+  // → a no-op.
+  bill: ({ db, user, model, usage, tools }) => (user == null ? void 0 : user.id) ? () => billRequest({ db, user, model, usage, tools }) : () => Promise.resolve()
+};
+
+// lib/ai.js
+var google_client = new GoogleGenAI({
+  apiKey: process.env.GOOGLE_GEMINI_API_KEY
+});
+var googleCircuit = circuit({
+  name: "google-ai",
+  threshold: 3,
+  timeout: 6e4
+});
+var models = {
+  image: "gemini-2.5-flash-image",
+  text: "gemini-2.5-flash"
+};
+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,
+  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 google = {
   image: async ({
     config = {},
@@ -267,7 +275,7 @@ var google = {
     }
     const usage = tokensFromMetadata(response == null ? void 0 : response.usageMetadata);
     const tools = toolsFromResponse(response);
-    const bill = (user == null ? void 0 : user.id) ? () => billRequest({ db, user, model, usage, tools }) : () => Promise.resolve();
+    const bill = ai.bill({ db, user, model, usage, tools });
     return { content, bill };
   },
   text: async ({
@@ -295,17 +303,11 @@ var google = {
     }
     const usage = tokensFromMetadata(response == null ? void 0 : response.usageMetadata);
     const tools = toolsFromResponse(response);
-    const bill = (user == null ? void 0 : user.id) ? () => billRequest({ db, user, model, usage, tools }) : () => Promise.resolve();
+    const bill = ai.bill({ db, user, model, usage, tools });
     return { content, bill };
   }
 };
 export {
   MARKUP,
-  cost,
-  costForRequest,
-  google,
-  priceForRequest,
-  pricing,
-  toolCost,
-  toolPricing
+  google
 };