@drawbridge/drawbridge-utils 0.0.79 → 0.0.80

package/dist/ai.cjs

@@ -1,6 +1,8 @@
+var __create = Object.create;
 var __defProp = Object.defineProperty;
 var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
 var __getOwnPropNames = Object.getOwnPropertyNames;
+var __getProtoOf = Object.getPrototypeOf;
 var __hasOwnProp = Object.prototype.hasOwnProperty;
 var __export = (target, all) => {
   for (var name in all)
@@ -14,6 +16,14 @@ var __copyProps = (to, from, except, desc) => {
   }
   return to;
 };
+var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(
+  // If the importer is in node compatibility mode or this is not an ESM
+  // file that has been converted to a CommonJS file using a Babel-
+  // compatible transform (i.e. "__esModule" has not been set), then set
+  // "default" to the CommonJS "module.exports" for node compatibility.
+  isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target,
+  mod
+));
 var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
 
 // lib/ai.js
@@ -251,6 +261,13 @@ var ai = {
   bill: ({ db, user, model, usage, tools }) => (user == null ? void 0 : user.id) ? () => billRequest({ db, user, model, usage, tools }) : () => Promise.resolve()
 };
 
+// lib/fetch.js
+var import_qs = __toESM(require("qs"), 1);
+var isTransientError = (error) => {
+  const status = error == null ? void 0 : error.status;
+  return !status || status === 429 || status >= 500;
+};
+
 // lib/ai.js
 var google_client = new import_genai.GoogleGenAI({
   apiKey: process.env.GOOGLE_GEMINI_API_KEY
@@ -260,6 +277,16 @@ var googleCircuit = circuit({
   threshold: 3,
   timeout: 6e4
 });
+var withRetries = async (fn, { attempts = 3, delay = 500 } = {}) => {
+  for (let attempt = 1; ; attempt++) {
+    try {
+      return await fn();
+    } catch (error) {
+      if (attempt >= attempts || !isTransientError(error)) throw error;
+      await new Promise((resolve) => setTimeout(resolve, delay * attempt));
+    }
+  }
+};
 var models = {
   image: "gemini-2.5-flash-image",
   text: "gemini-3.5-flash"
@@ -293,14 +320,14 @@ var google = {
       ...refs.map((ref) => ({ inlineData: { data: ref.data, mimeType: ref.mimeType } })),
       { text: prompt.join("\n") }
     ];
-    const response = await googleCircuit(() => google_client.models.generateContent({
+    const response = await googleCircuit(() => withRetries(() => google_client.models.generateContent({
       model,
       contents: { parts },
       config: {
         ...config,
         responseModalities: ["Image"]
       }
-    }));
+    })));
     const content = (_f = (_e = (_d = (_c = (_b = (_a = response == null ? void 0 : response.candidates) == null ? void 0 : _a[0]) == null ? void 0 : _b.content) == null ? void 0 : _c.parts) == null ? void 0 : _d[0]) == null ? void 0 : _e.inlineData) == null ? void 0 : _f.data;
     if (!content) {
       throw new Error("Could not generate from prompt. Please try again with a different prompt.");
@@ -321,14 +348,14 @@ var google = {
   }) => {
     const model = models.text;
     const contents = parts ? { parts } : images.length ? { parts: [...images.map((image) => ({ inlineData: { data: image.data, mimeType: image.mimeType } })), { text: prompt.join("\n") }] } : prompt.join("\n");
-    const response = await googleCircuit(() => google_client.models.generateContent({
+    const response = await googleCircuit(() => withRetries(() => google_client.models.generateContent({
       config: {
         ...config,
         responseModalities: ["Text"]
       },
       contents,
       model
-    }));
+    })));
     const content = response == null ? void 0 : response.text;
     if (!content) {
       throw new Error("Could not generate from prompt. Please try again with a different prompt.");

package/dist/ai.d.cts

@@ -2,9 +2,11 @@ import { GoogleGenAI } from '@google/genai';
 import { circuit } from './circuit.cjs';
 import { ai } from './billing.cjs';
 export { MARKUP } from './billing.cjs';
+import { isTransientError } from './fetch.cjs';
 import './transactions.cjs';
 import '@drawbridge/drawbridge-telemetry';
 import './usage.cjs';
+import 'qs';
 
 // AI request entry point. Wraps the Google GenAI client with a per-process
 // circuit breaker, owns the per-model pricing table, and atomically debits
@@ -35,6 +37,35 @@ const googleCircuit = circuit({
 	timeout : 60000
 });
 
+// Bounded retry for the raw model call. A single transient failure — undici
+// "fetch failed" on a resource-starved pod, a 429, a 5xx — must not lose the
+// generation step OR count toward the breaker: three of those trip it and
+// hard-fail every AI call for the next 60s. Runs INSIDE the googleCircuit
+// callback so the breaker only counts calls that failed after retries were
+// exhausted, and an OPEN circuit still fails fast without retrying.
+// isTransientError treats status-less errors as transient, which is correct
+// here: the wrapped fn is exactly one SDK call, and the GenAI SDK reports
+// network-level failures as plain Errors with no `.status`.
+const withRetries = async ( fn, { attempts = 3, delay = 500 } = {} ) => {
+
+	for( let attempt = 1; ; attempt++ ){
+
+		try {
+
+			return await fn();
+
+		} catch ( error ) {
+
+			if( attempt >= attempts || ! isTransientError( error ) ) throw error;
+
+			await new Promise( ( resolve ) => setTimeout( resolve, delay * attempt ) );
+
+		}
+
+	}
+
+};
+
 // gemini-2.5-flash was retired by Google (generateContent → 404 "no longer
 // available"), taking down every generation. gemini-3.5-flash is the current
 // stable flash successor. The image model is a distinct model and still live,
@@ -89,14 +120,14 @@ const google = {
 			{ text : prompt.join( '\n' ) }
 		];
 
-		const response = await googleCircuit( () => google_client.models.generateContent({
+		const response = await googleCircuit( () => withRetries( () => google_client.models.generateContent({
 			model,
 			contents : { parts },
 			config : {
 				...config,
 				responseModalities : [ 'Image' ]
 			}
-		}) );
+		}) ) );
 
 		const content = response?.candidates?.[ 0 ]?.content?.parts?.[ 0 ]?.inlineData?.data;
 
@@ -143,14 +174,14 @@ const google = {
 				? { parts : [ ...images.map( ( image ) => ({ inlineData : { data : image.data, mimeType : image.mimeType } }) ), { text : prompt.join( '\n' ) } ] }
 				: prompt.join( '\n' );
 
-		const response = await googleCircuit( () => google_client.models.generateContent({
+		const response = await googleCircuit( () => withRetries( () => google_client.models.generateContent({
 			config : {
 				...config,
 				responseModalities : [ 'Text' ]
 			},
 			contents,
 			model
-		}) );
+		}) ) );
 
 		const content = response?.text;
 

package/dist/ai.d.ts

@@ -2,9 +2,11 @@ import { GoogleGenAI } from '@google/genai';
 import { circuit } from './circuit.js';
 import { ai } from './billing.js';
 export { MARKUP } from './billing.js';
+import { isTransientError } from './fetch.js';
 import './transactions.js';
 import '@drawbridge/drawbridge-telemetry';
 import './usage.js';
+import 'qs';
 
 // AI request entry point. Wraps the Google GenAI client with a per-process
 // circuit breaker, owns the per-model pricing table, and atomically debits
@@ -35,6 +37,35 @@ const googleCircuit = circuit({
 	timeout : 60000
 });
 
+// Bounded retry for the raw model call. A single transient failure — undici
+// "fetch failed" on a resource-starved pod, a 429, a 5xx — must not lose the
+// generation step OR count toward the breaker: three of those trip it and
+// hard-fail every AI call for the next 60s. Runs INSIDE the googleCircuit
+// callback so the breaker only counts calls that failed after retries were
+// exhausted, and an OPEN circuit still fails fast without retrying.
+// isTransientError treats status-less errors as transient, which is correct
+// here: the wrapped fn is exactly one SDK call, and the GenAI SDK reports
+// network-level failures as plain Errors with no `.status`.
+const withRetries = async ( fn, { attempts = 3, delay = 500 } = {} ) => {
+
+	for( let attempt = 1; ; attempt++ ){
+
+		try {
+
+			return await fn();
+
+		} catch ( error ) {
+
+			if( attempt >= attempts || ! isTransientError( error ) ) throw error;
+
+			await new Promise( ( resolve ) => setTimeout( resolve, delay * attempt ) );
+
+		}
+
+	}
+
+};
+
 // gemini-2.5-flash was retired by Google (generateContent → 404 "no longer
 // available"), taking down every generation. gemini-3.5-flash is the current
 // stable flash successor. The image model is a distinct model and still live,
@@ -89,14 +120,14 @@ const google = {
 			{ text : prompt.join( '\n' ) }
 		];
 
-		const response = await googleCircuit( () => google_client.models.generateContent({
+		const response = await googleCircuit( () => withRetries( () => google_client.models.generateContent({
 			model,
 			contents : { parts },
 			config : {
 				...config,
 				responseModalities : [ 'Image' ]
 			}
-		}) );
+		}) ) );
 
 		const content = response?.candidates?.[ 0 ]?.content?.parts?.[ 0 ]?.inlineData?.data;
 
@@ -143,14 +174,14 @@ const google = {
 				? { parts : [ ...images.map( ( image ) => ({ inlineData : { data : image.data, mimeType : image.mimeType } }) ), { text : prompt.join( '\n' ) } ] }
 				: prompt.join( '\n' );
 
-		const response = await googleCircuit( () => google_client.models.generateContent({
+		const response = await googleCircuit( () => withRetries( () => google_client.models.generateContent({
 			config : {
 				...config,
 				responseModalities : [ 'Text' ]
 			},
 			contents,
 			model
-		}) );
+		}) ) );
 
 		const content = response?.text;
 

package/dist/ai.js

@@ -227,6 +227,13 @@ var ai = {
   bill: ({ db, user, model, usage, tools }) => (user == null ? void 0 : user.id) ? () => billRequest({ db, user, model, usage, tools }) : () => Promise.resolve()
 };
 
+// lib/fetch.js
+import qs from "qs";
+var isTransientError = (error) => {
+  const status = error == null ? void 0 : error.status;
+  return !status || status === 429 || status >= 500;
+};
+
 // lib/ai.js
 var google_client = new GoogleGenAI({
   apiKey: process.env.GOOGLE_GEMINI_API_KEY
@@ -236,6 +243,16 @@ var googleCircuit = circuit({
   threshold: 3,
   timeout: 6e4
 });
+var withRetries = async (fn, { attempts = 3, delay = 500 } = {}) => {
+  for (let attempt = 1; ; attempt++) {
+    try {
+      return await fn();
+    } catch (error) {
+      if (attempt >= attempts || !isTransientError(error)) throw error;
+      await new Promise((resolve) => setTimeout(resolve, delay * attempt));
+    }
+  }
+};
 var models = {
   image: "gemini-2.5-flash-image",
   text: "gemini-3.5-flash"
@@ -269,14 +286,14 @@ var google = {
       ...refs.map((ref) => ({ inlineData: { data: ref.data, mimeType: ref.mimeType } })),
       { text: prompt.join("\n") }
     ];
-    const response = await googleCircuit(() => google_client.models.generateContent({
+    const response = await googleCircuit(() => withRetries(() => google_client.models.generateContent({
       model,
       contents: { parts },
       config: {
         ...config,
         responseModalities: ["Image"]
       }
-    }));
+    })));
     const content = (_f = (_e = (_d = (_c = (_b = (_a = response == null ? void 0 : response.candidates) == null ? void 0 : _a[0]) == null ? void 0 : _b.content) == null ? void 0 : _c.parts) == null ? void 0 : _d[0]) == null ? void 0 : _e.inlineData) == null ? void 0 : _f.data;
     if (!content) {
       throw new Error("Could not generate from prompt. Please try again with a different prompt.");
@@ -297,14 +314,14 @@ var google = {
   }) => {
     const model = models.text;
     const contents = parts ? { parts } : images.length ? { parts: [...images.map((image) => ({ inlineData: { data: image.data, mimeType: image.mimeType } })), { text: prompt.join("\n") }] } : prompt.join("\n");
-    const response = await googleCircuit(() => google_client.models.generateContent({
+    const response = await googleCircuit(() => withRetries(() => google_client.models.generateContent({
       config: {
         ...config,
         responseModalities: ["Text"]
       },
       contents,
       model
-    }));
+    })));
     const content = response == null ? void 0 : response.text;
     if (!content) {
       throw new Error("Could not generate from prompt. Please try again with a different prompt.");

package/package.json

@@ -2,7 +2,7 @@
   "type": "module",
   "dependencies": {
     "@drawbridge/drawbridge-agents": "0.0.10",
-    "@drawbridge/drawbridge-telemetry": "0.0.14",
+    "@drawbridge/drawbridge-telemetry": "0.0.15",
     "@google/genai": "1.30.0",
     "axios": "1.16.0",
     "currency-codes": "2.2.0",
@@ -164,5 +164,5 @@
     "test": "node --test test/"
   },
   "types": "dist/index.d.ts",
-  "version": "0.0.79"
+  "version": "0.0.80"
 }