@falai/agent 2.4.3 → 2.5.0

package/dist/utils/retry.d.ts

@@ -5,8 +5,37 @@ export interface RetryOptions<T> {
     onRetry?: (attempt: number, error: unknown) => void;
     onFailure?: (error: unknown) => boolean;
 }
+/** Provider timeout (ms) + retry count, after defaults are applied. */
+export interface RetryConfig {
+    timeout: number;
+    retries: number;
+}
+/**
+ * Normalize a provider's optional retry config against the defaults. `timeout`
+ * uses `||` (a 0ms timeout is degenerate — it aborts every call immediately — so
+ * fall back to the default), while `retries` uses `??` so an explicit
+ * `retries: 0` (disable retries) is honored rather than treated as unset. Single
+ * definition so this distinction can't drift between providers.
+ */
+export declare function resolveRetryConfig(input?: {
+    timeout?: number;
+    retries?: number;
+}): RetryConfig;
 export declare function retry<T>(options: RetryOptions<T>): Promise<T>;
-export declare const withTimeoutAndRetry: <T>(operation: () => Promise<T>, timeoutMs?: number, maxRetries?: number, operationName?: string) => Promise<T>;
+/**
+ * Combine abort signals into one that aborts as soon as any input aborts.
+ * `undefined` inputs are skipped, so a caller can pass an optional caller-signal
+ * alongside a required per-attempt one. Returns the sole signal when only one is
+ * present (no wrapper), or `undefined` when none are.
+ *
+ * Prefers the platform `AbortSignal.any`, which cleans up its listeners via weak
+ * refs — so merging onto a long-lived caller signal that's reused across many
+ * calls can't accumulate listeners. Falls back to a manual controller on older
+ * runtimes (< Node 20.3); there the listeners live until a source aborts or is
+ * garbage-collected, which is fine for the usual per-request signal.
+ */
+export declare function combineAbortSignals(...signals: Array<AbortSignal | undefined>): AbortSignal | undefined;
+export declare const withTimeoutAndRetry: <T>(operation: (signal: AbortSignal) => Promise<T>, timeoutMs?: number, maxRetries?: number, operationName?: string) => Promise<T>;
 export interface StreamRetryOptions {
     /** Maximum number of retries after the first attempt. Defaults to 3. */
     maxRetries?: number;
@@ -14,18 +43,25 @@ export interface StreamRetryOptions {
     delay?: (attempt: number) => number;
     /** Label used in retry logs. */
     operationName?: string;
+    /**
+     * Max ms to wait for the *first* chunk (time-to-first-token) before treating
+     * the attempt as failed. Guards a provider that opens a stream and then
+     * stalls. Only the first chunk is bounded — later chunks are unbounded so a
+     * long but healthy stream is never cut off. `0`/omitted disables the deadline.
+     */
+    firstChunkTimeoutMs?: number;
 }
 /**
  * Streaming analog of {@link withTimeoutAndRetry}. Re-runs an async-generator
- * factory as long as it fails *before yielding its first chunk* — e.g. an
- * empty completion that throws "No response", or an error while establishing
- * the stream. Once any chunk has been yielded the stream is committed and
- * further errors propagate, so a retry can never double-emit deltas that the
- * consumer has already received.
+ * factory as long as it fails *before yielding its first chunk* — an empty
+ * completion that throws "No response", an error while establishing the stream,
+ * or (with `firstChunkTimeoutMs`) a stall before the first token. Once any chunk
+ * has been yielded the stream is committed and further errors propagate, so a
+ * retry can never double-emit deltas the consumer has already received.
  *
  * This mirrors the non-streaming path, where the provider throws on an empty
  * completion inside `withTimeoutAndRetry` and is retried on the same model
  * before the caller falls through to backup models.
  */
-export declare function withStreamRetry<T>(factory: () => AsyncGenerator<T>, options?: StreamRetryOptions): AsyncGenerator<T>;
+export declare function withStreamRetry<T>(factory: (signal: AbortSignal) => AsyncGenerator<T>, options?: StreamRetryOptions): AsyncGenerator<T>;
 //# sourceMappingURL=retry.d.ts.map

package/dist/utils/retry.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"retry.d.ts","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAKA,MAAM,WAAW,YAAY,CAAC,CAAC;IAC7B,SAAS,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,CAAC;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,MAAM,CAAC;IACnC,OAAO,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,KAAK,IAAI,CAAC;IACpD,SAAS,CAAC,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,OAAO,CAAC;CACzC;AAED,wBAAsB,KAAK,CAAC,CAAC,EAAE,OAAO,EAAE,YAAY,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CA8BnE;AAED,eAAO,MAAM,mBAAmB,GAAU,CAAC,EACzC,WAAW,MAAM,OAAO,CAAC,CAAC,CAAC,EAC3B,YAAW,MAAc,EACzB,aAAY,MAAU,EACtB,gBAAe,MAAuB,KACrC,OAAO,CAAC,CAAC,CA8CX,CAAC;AAEF,MAAM,WAAW,kBAAkB;IACjC,wEAAwE;IACxE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,8EAA8E;IAC9E,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,MAAM,CAAC;IACpC,gCAAgC;IAChC,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED;;;;;;;;;;;GAWG;AACH,wBAAuB,eAAe,CAAC,CAAC,EACtC,OAAO,EAAE,MAAM,cAAc,CAAC,CAAC,CAAC,EAChC,OAAO,GAAE,kBAAuB,GAC/B,cAAc,CAAC,CAAC,CAAC,CA6BnB"}
+{"version":3,"file":"retry.d.ts","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAKA,MAAM,WAAW,YAAY,CAAC,CAAC;IAC7B,SAAS,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,CAAC;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,MAAM,CAAC;IACnC,OAAO,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,KAAK,IAAI,CAAC;IACpD,SAAS,CAAC,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,OAAO,CAAC;CACzC;AAED,uEAAuE;AACvE,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;CACjB;AAOD;;;;;;GAMG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,CAAC,EAAE;IACzC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,GAAG,WAAW,CAKd;AAED,wBAAsB,KAAK,CAAC,CAAC,EAAE,OAAO,EAAE,YAAY,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CA8BnE;AAMD;;;;;;;;;;;GAWG;AACH,wBAAgB,mBAAmB,CACjC,GAAG,OAAO,EAAE,KAAK,CAAC,WAAW,GAAG,SAAS,CAAC,GACzC,WAAW,GAAG,SAAS,CAgBzB;AAED,eAAO,MAAM,mBAAmB,GAAU,CAAC,EACzC,WAAW,CAAC,MAAM,EAAE,WAAW,KAAK,OAAO,CAAC,CAAC,CAAC,EAC9C,YAAW,MAAc,EACzB,aAAY,MAAU,EACtB,gBAAe,MAAuB,KACrC,OAAO,CAAC,CAAC,CAiDX,CAAC;AAEF,MAAM,WAAW,kBAAkB;IACjC,wEAAwE;IACxE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,8EAA8E;IAC9E,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,MAAM,CAAC;IACpC,gCAAgC;IAChC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB;;;;;OAKG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAC;CAC9B;AAuBD;;;;;;;;;;;GAWG;AACH,wBAAuB,eAAe,CAAC,CAAC,EACtC,OAAO,EAAE,CAAC,MAAM,EAAE,WAAW,KAAK,cAAc,CAAC,CAAC,CAAC,EACnD,OAAO,GAAE,kBAAuB,GAC/B,cAAc,CAAC,CAAC,CAAC,CAwDnB"}

package/dist/utils/retry.js

@@ -2,6 +2,23 @@
  * Retry utility with exponential backoff
  */
 import { logger } from './logger';
+const DEFAULT_RETRY_CONFIG = {
+    timeout: 60000,
+    retries: 3,
+};
+/**
+ * Normalize a provider's optional retry config against the defaults. `timeout`
+ * uses `||` (a 0ms timeout is degenerate — it aborts every call immediately — so
+ * fall back to the default), while `retries` uses `??` so an explicit
+ * `retries: 0` (disable retries) is honored rather than treated as unset. Single
+ * definition so this distinction can't drift between providers.
+ */
+export function resolveRetryConfig(input) {
+    return {
+        timeout: input?.timeout || DEFAULT_RETRY_CONFIG.timeout,
+        retries: input?.retries ?? DEFAULT_RETRY_CONFIG.retries,
+    };
+}
 export async function retry(options) {
     const { operation, maxRetries, delay, onRetry, onFailure } = options;
     let lastError;
@@ -28,6 +45,38 @@ export async function retry(options) {
     // eslint-disable-next-line @typescript-eslint/only-throw-error
     throw lastError ?? new Error("Operation failed");
 }
+/** Capped exponential backoff (1s, 2s, 4s, … max 5s) shared by the retry helpers. */
+const defaultBackoff = (attempt) => Math.min(1000 * Math.pow(2, attempt), 5000);
+/**
+ * Combine abort signals into one that aborts as soon as any input aborts.
+ * `undefined` inputs are skipped, so a caller can pass an optional caller-signal
+ * alongside a required per-attempt one. Returns the sole signal when only one is
+ * present (no wrapper), or `undefined` when none are.
+ *
+ * Prefers the platform `AbortSignal.any`, which cleans up its listeners via weak
+ * refs — so merging onto a long-lived caller signal that's reused across many
+ * calls can't accumulate listeners. Falls back to a manual controller on older
+ * runtimes (< Node 20.3); there the listeners live until a source aborts or is
+ * garbage-collected, which is fine for the usual per-request signal.
+ */
+export function combineAbortSignals(...signals) {
+    const present = signals.filter((s) => s != null);
+    if (present.length <= 1)
+        return present[0];
+    if (typeof AbortSignal.any === "function")
+        return AbortSignal.any(present);
+    const controller = new AbortController();
+    for (const signal of present) {
+        if (signal.aborted) {
+            controller.abort(signal.reason);
+            break;
+        }
+        signal.addEventListener("abort", () => controller.abort(signal.reason), {
+            once: true,
+        });
+    }
+    return controller.signal;
+}
 export const withTimeoutAndRetry = async (operation, timeoutMs = 60000, maxRetries = 3, operationName = "AI operation") => {
     const createTimeoutOperation = () => async () => {
         const controller = new AbortController();
@@ -35,8 +84,11 @@ export const withTimeoutAndRetry = async (operation, timeoutMs = 60000, maxRetri
             controller.abort();
         }, timeoutMs);
         try {
+            // Hand the operation the timeout signal so the in-flight upstream call is
+            // actually cancelled when the deadline fires — otherwise the abandoned
+            // attempt keeps running while the retry stacks a second concurrent call.
             const result = await Promise.race([
-                operation(),
+                operation(controller.signal),
                 new Promise((_, reject) => {
                     controller.signal.addEventListener("abort", () => {
                         reject(new Error(`Operation timed out after ${timeoutMs}ms`));
@@ -54,11 +106,11 @@ export const withTimeoutAndRetry = async (operation, timeoutMs = 60000, maxRetri
     return retry({
         operation: createTimeoutOperation(),
         maxRetries,
-        delay: (attempt) => Math.min(1000 * Math.pow(2, attempt), 5000),
+        delay: defaultBackoff,
         onRetry: (attempt, error) => {
             const message = error instanceof Error ? error.message : String(error);
             logger.error(`[${operationName}] Failed attempt ${attempt + 1}:`, message);
-            const delay = Math.min(1000 * Math.pow(2, attempt), 5000);
+            const delay = defaultBackoff(attempt);
             logger.debug(`[${operationName}] Retrying in ${delay}ms...`);
             logger.debug(`[${operationName}] Attempt ${attempt + 2}/${maxRetries + 1}`);
         },
@@ -68,28 +120,49 @@ export const withTimeoutAndRetry = async (operation, timeoutMs = 60000, maxRetri
         },
     });
 };
+/** Reject if `next` hasn't settled within `timeoutMs`; always clears its timer. */
+function raceFirstChunk(next, timeoutMs, operationName) {
+    let timer;
+    const timeout = new Promise((_, reject) => {
+        timer = setTimeout(() => reject(new Error(`[${operationName}] Stream timed out: no first chunk within ${timeoutMs}ms`)), timeoutMs);
+    });
+    return Promise.race([next, timeout]).finally(() => clearTimeout(timer));
+}
 /**
  * Streaming analog of {@link withTimeoutAndRetry}. Re-runs an async-generator
- * factory as long as it fails *before yielding its first chunk* — e.g. an
- * empty completion that throws "No response", or an error while establishing
- * the stream. Once any chunk has been yielded the stream is committed and
- * further errors propagate, so a retry can never double-emit deltas that the
- * consumer has already received.
+ * factory as long as it fails *before yielding its first chunk* — an empty
+ * completion that throws "No response", an error while establishing the stream,
+ * or (with `firstChunkTimeoutMs`) a stall before the first token. Once any chunk
+ * has been yielded the stream is committed and further errors propagate, so a
+ * retry can never double-emit deltas the consumer has already received.
  *
  * This mirrors the non-streaming path, where the provider throws on an empty
  * completion inside `withTimeoutAndRetry` and is retried on the same model
  * before the caller falls through to backup models.
  */
 export async function* withStreamRetry(factory, options = {}) {
-    const { maxRetries = 3, delay = (attempt) => Math.min(1000 * Math.pow(2, attempt), 5000), operationName = "AI stream", } = options;
+    const { maxRetries = 3, delay = defaultBackoff, operationName = "AI stream", firstChunkTimeoutMs, } = options;
     for (let attempt = 0; attempt <= maxRetries; attempt++) {
+        // One controller per attempt: aborting it on abandon cancels the upstream
+        // SDK call so a retry can't stack a second concurrent stream.
+        const controller = new AbortController();
+        const iterator = factory(controller.signal);
         let yielded = false;
+        let completed = false;
         try {
-            for await (const chunk of factory()) {
+            // Drive the iterator by hand so the first chunk can be raced against a
+            // deadline; for-await would give no hook for that.
+            for (;;) {
+                const result = !yielded && firstChunkTimeoutMs
+                    ? await raceFirstChunk(iterator.next(), firstChunkTimeoutMs, operationName)
+                    : await iterator.next();
+                if (result.done) {
+                    completed = true;
+                    return;
+                }
                 yielded = true;
-                yield chunk;
+                yield result.value;
             }
-            return;
         }
         catch (error) {
             // Can't retry once deltas are out, and don't retry past the budget.
@@ -102,6 +175,19 @@ export async function* withStreamRetry(factory, options = {}) {
             logger.debug(`[${operationName}] Retrying in ${delayMs}ms... (attempt ${attempt + 2}/${maxRetries + 1})`);
             await new Promise((resolve) => setTimeout(resolve, delayMs));
         }
+        finally {
+            // Abandoning an attempt (first-chunk timeout, pre-yield error before a
+            // retry, or a consumer that breaks early): abort the upstream SDK call
+            // first so it's actually torn down — not left running while we retry —
+            // then dispose the iterator. Aborting also unblocks a stalled
+            // iterator.next() so return() can run the generator's cleanup. Both are
+            // fire-and-forget: a wedged generator's return() may never settle, so we
+            // must not await it.
+            if (!completed) {
+                controller.abort();
+                void iterator.return?.(undefined)?.catch(() => undefined);
+            }
+        }
     }
 }
 //# sourceMappingURL=retry.js.map

package/dist/utils/retry.js.map

@@ -1 +1 @@
-{"version":3,"file":"retry.js","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAC,MAAM,EAAC,MAAM,UAAU,CAAA;AAU/B,MAAM,CAAC,KAAK,UAAU,KAAK,CAAI,OAAwB;IACrD,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,OAAO,CAAC;IAErE,IAAI,SAAkB,CAAC;IAEvB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;QACvD,IAAI,CAAC;YACH,OAAO,MAAM,SAAS,EAAE,CAAC;QAC3B,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,SAAS,GAAG,KAAK,CAAC;YAElB,IAAI,OAAO,KAAK,UAAU,EAAE,CAAC;gBAC3B,MAAM,aAAa,GAAG,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;gBAC9D,IAAI,aAAa,EAAE,CAAC;oBAClB,MAAM,SAAS,CAAC;gBAClB,CAAC;gBACD,MAAM;YACR,CAAC;YAED,IAAI,OAAO,EAAE,CAAC;gBACZ,OAAO,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;YAC9B,CAAC;YAED,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC;YAC/B,MAAM,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC;QAC/D,CAAC;IACH,CAAC;IAED,+DAA+D;IAC/D,MAAM,SAAS,IAAI,IAAI,KAAK,CAAC,kBAAkB,CAAC,CAAC;AACnD,CAAC;AAED,MAAM,CAAC,MAAM,mBAAmB,GAAG,KAAK,EACtC,SAA2B,EAC3B,YAAoB,KAAK,EACzB,aAAqB,CAAC,EACtB,gBAAwB,cAAc,EAC1B,EAAE;IACd,MAAM,sBAAsB,GAAG,GAAG,EAAE,CAAC,KAAK,IAAgB,EAAE;QAC1D,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,SAAS,GAAG,UAAU,CAAC,GAAG,EAAE;YAChC,UAAU,CAAC,KAAK,EAAE,CAAC;QACrB,CAAC,EAAE,SAAS,CAAC,CAAC;QAEd,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC;gBAChC,SAAS,EAAE;gBACX,IAAI,OAAO,CAAQ,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE;oBAC/B,UAAU,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,GAAG,EAAE;wBAC/C,MAAM,CAAC,IAAI,KAAK,CAAC,6BAA6B,SAAS,IAAI,CAAC,CAAC,CAAC;oBAChE,CAAC,CAAC,CAAC;gBACL,CAAC,CAAC;aACH,CAAC,CAAC;YAEH,YAAY,CAAC,SAAS,CAAC,CAAC;YACxB,OAAO,MAAM,CAAC;QAChB,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,YAAY,CAAC,SAAS,CAAC,CAAC;YACxB,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC,CAAC;IAEF,OAAO,KAAK,CAAI;QACd,SAAS,EAAE,sBAAsB,EAAE;QACnC,UAAU;QACV,KAAK,EAAE,CAAC,OAAe,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,EAAE,IAAI,CAAC;QACvE,OAAO,EAAE,CAAC,OAAe,EAAE,KAAc,EAAE,EAAE;YAC3C,MAAM,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YACvE,MAAM,CAAC,KAAK,CACV,IAAI,aAAa,oBAAoB,OAAO,GAAG,CAAC,GAAG,EACnD,OAAO,CACR,CAAC;YACF,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,EAAE,IAAI,CAAC,CAAC;YAC1D,MAAM,CAAC,KAAK,CAAC,IAAI,aAAa,iBAAiB,KAAK,OAAO,CAAC,CAAC;YAC7D,MAAM,CAAC,KAAK,CACV,IAAI,aAAa,aAAa,OAAO,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,EAAE,CAC9D,CAAC;QACJ,CAAC;QACD,SAAS,EAAE,CAAC,MAAe,EAAE,EAAE;YAC7B,MAAM,CAAC,KAAK,CAAC,IAAI,aAAa,SAAS,UAAU,GAAG,CAAC,kBAAkB,CAAC,CAAC;YACzE,OAAO,IAAI,CAAC;QACd,CAAC;KACF,CAAC,CAAC;AACL,CAAC,CAAC;AAWF;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,KAAK,SAAS,CAAC,CAAC,eAAe,CACpC,OAAgC,EAChC,UAA8B,EAAE;IAEhC,MAAM,EACJ,UAAU,GAAG,CAAC,EACd,KAAK,GAAG,CAAC,OAAe,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,EAAE,IAAI,CAAC,EACxE,aAAa,GAAG,WAAW,GAC5B,GAAG,OAAO,CAAC;IAEZ,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;QACvD,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,IAAI,CAAC;YACH,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,OAAO,EAAE,EAAE,CAAC;gBACpC,OAAO,GAAG,IAAI,CAAC;gBACf,MAAM,KAAK,CAAC;YACd,CAAC;YACD,OAAO;QACT,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,oEAAoE;YACpE,IAAI,OAAO,IAAI,OAAO,KAAK,UAAU,EAAE,CAAC;gBACtC,MAAM,KAAK,CAAC;YACd,CAAC;YACD,MAAM,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YACvE,MAAM,CAAC,KAAK,CAAC,IAAI,aAAa,oBAAoB,OAAO,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;YAC3E,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC;YAC/B,MAAM,CAAC,KAAK,CACV,IAAI,aAAa,iBAAiB,OAAO,kBAAkB,OAAO,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,GAAG,CAC5F,CAAC;YACF,MAAM,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC;QAC/D,CAAC;IACH,CAAC;AACH,CAAC"}
+{"version":3,"file":"retry.js","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAC,MAAM,EAAC,MAAM,UAAU,CAAA;AAgB/B,MAAM,oBAAoB,GAAgB;IACxC,OAAO,EAAE,KAAK;IACd,OAAO,EAAE,CAAC;CACX,CAAC;AAEF;;;;;;GAMG;AACH,MAAM,UAAU,kBAAkB,CAAC,KAGlC;IACC,OAAO;QACL,OAAO,EAAE,KAAK,EAAE,OAAO,IAAI,oBAAoB,CAAC,OAAO;QACvD,OAAO,EAAE,KAAK,EAAE,OAAO,IAAI,oBAAoB,CAAC,OAAO;KACxD,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,KAAK,CAAI,OAAwB;IACrD,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,OAAO,CAAC;IAErE,IAAI,SAAkB,CAAC;IAEvB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;QACvD,IAAI,CAAC;YACH,OAAO,MAAM,SAAS,EAAE,CAAC;QAC3B,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,SAAS,GAAG,KAAK,CAAC;YAElB,IAAI,OAAO,KAAK,UAAU,EAAE,CAAC;gBAC3B,MAAM,aAAa,GAAG,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;gBAC9D,IAAI,aAAa,EAAE,CAAC;oBAClB,MAAM,SAAS,CAAC;gBAClB,CAAC;gBACD,MAAM;YACR,CAAC;YAED,IAAI,OAAO,EAAE,CAAC;gBACZ,OAAO,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;YAC9B,CAAC;YAED,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC;YAC/B,MAAM,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC;QAC/D,CAAC;IACH,CAAC;IAED,+DAA+D;IAC/D,MAAM,SAAS,IAAI,IAAI,KAAK,CAAC,kBAAkB,CAAC,CAAC;AACnD,CAAC;AAED,qFAAqF;AACrF,MAAM,cAAc,GAAG,CAAC,OAAe,EAAU,EAAE,CACjD,IAAI,CAAC,GAAG,CAAC,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,EAAE,IAAI,CAAC,CAAC;AAE9C;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,mBAAmB,CACjC,GAAG,OAAuC;IAE1C,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAoB,EAAE,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC;IACnE,IAAI,OAAO,CAAC,MAAM,IAAI,CAAC;QAAE,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC;IAC3C,IAAI,OAAO,WAAW,CAAC,GAAG,KAAK,UAAU;QAAE,OAAO,WAAW,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IAE3E,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YACnB,UAAU,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;YAChC,MAAM;QACR,CAAC;QACD,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE;YACtE,IAAI,EAAE,IAAI;SACX,CAAC,CAAC;IACL,CAAC;IACD,OAAO,UAAU,CAAC,MAAM,CAAC;AAC3B,CAAC;AAED,MAAM,CAAC,MAAM,mBAAmB,GAAG,KAAK,EACtC,SAA8C,EAC9C,YAAoB,KAAK,EACzB,aAAqB,CAAC,EACtB,gBAAwB,cAAc,EAC1B,EAAE;IACd,MAAM,sBAAsB,GAAG,GAAG,EAAE,CAAC,KAAK,IAAgB,EAAE;QAC1D,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,SAAS,GAAG,UAAU,CAAC,GAAG,EAAE;YAChC,UAAU,CAAC,KAAK,EAAE,CAAC;QACrB,CAAC,EAAE,SAAS,CAAC,CAAC;QAEd,IAAI,CAAC;YACH,0EAA0E;YAC1E,uEAAuE;YACvE,yEAAyE;YACzE,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC;gBAChC,SAAS,CAAC,UAAU,CAAC,MAAM,CAAC;gBAC5B,IAAI,OAAO,CAAQ,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE;oBAC/B,UAAU,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,GAAG,EAAE;wBAC/C,MAAM,CAAC,IAAI,KAAK,CAAC,6BAA6B,SAAS,IAAI,CAAC,CAAC,CAAC;oBAChE,CAAC,CAAC,CAAC;gBACL,CAAC,CAAC;aACH,CAAC,CAAC;YAEH,YAAY,CAAC,SAAS,CAAC,CAAC;YACxB,OAAO,MAAM,CAAC;QAChB,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,YAAY,CAAC,SAAS,CAAC,CAAC;YACxB,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC,CAAC;IAEF,OAAO,KAAK,CAAI;QACd,SAAS,EAAE,sBAAsB,EAAE;QACnC,UAAU;QACV,KAAK,EAAE,cAAc;QACrB,OAAO,EAAE,CAAC,OAAe,EAAE,KAAc,EAAE,EAAE;YAC3C,MAAM,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YACvE,MAAM,CAAC,KAAK,CACV,IAAI,aAAa,oBAAoB,OAAO,GAAG,CAAC,GAAG,EACnD,OAAO,CACR,CAAC;YACF,MAAM,KAAK,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;YACtC,MAAM,CAAC,KAAK,CAAC,IAAI,aAAa,iBAAiB,KAAK,OAAO,CAAC,CAAC;YAC7D,MAAM,CAAC,KAAK,CACV,IAAI,aAAa,aAAa,OAAO,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,EAAE,CAC9D,CAAC;QACJ,CAAC;QACD,SAAS,EAAE,CAAC,MAAe,EAAE,EAAE;YAC7B,MAAM,CAAC,KAAK,CAAC,IAAI,aAAa,SAAS,UAAU,GAAG,CAAC,kBAAkB,CAAC,CAAC;YACzE,OAAO,IAAI,CAAC;QACd,CAAC;KACF,CAAC,CAAC;AACL,CAAC,CAAC;AAkBF,mFAAmF;AACnF,SAAS,cAAc,CACrB,IAAgC,EAChC,SAAiB,EACjB,aAAqB;IAErB,IAAI,KAAgD,CAAC;IACrD,MAAM,OAAO,GAAG,IAAI,OAAO,CAAQ,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE;QAC/C,KAAK,GAAG,UAAU,CAChB,GAAG,EAAE,CACH,MAAM,CACJ,IAAI,KAAK,CACP,IAAI,aAAa,6CAA6C,SAAS,IAAI,CAC5E,CACF,EACH,SAAS,CACV,CAAC;IACJ,CAAC,CAAC,CAAC;IACH,OAAO,OAAO,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC;AAC1E,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,KAAK,SAAS,CAAC,CAAC,eAAe,CACpC,OAAmD,EACnD,UAA8B,EAAE;IAEhC,MAAM,EACJ,UAAU,GAAG,CAAC,EACd,KAAK,GAAG,cAAc,EACtB,aAAa,GAAG,WAAW,EAC3B,mBAAmB,GACpB,GAAG,OAAO,CAAC;IAEZ,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;QACvD,0EAA0E;QAC1E,8DAA8D;QAC9D,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,QAAQ,GAAG,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC;QAC5C,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,IAAI,SAAS,GAAG,KAAK,CAAC;QACtB,IAAI,CAAC;YACH,uEAAuE;YACvE,mDAAmD;YACnD,SAAS,CAAC;gBACR,MAAM,MAAM,GACV,CAAC,OAAO,IAAI,mBAAmB;oBAC7B,CAAC,CAAC,MAAM,cAAc,CAAC,QAAQ,CAAC,IAAI,EAAE,EAAE,mBAAmB,EAAE,aAAa,CAAC;oBAC3E,CAAC,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;gBAC5B,IAAI,MAAM,CAAC,IAAI,EAAE,CAAC;oBAChB,SAAS,GAAG,IAAI,CAAC;oBACjB,OAAO;gBACT,CAAC;gBACD,OAAO,GAAG,IAAI,CAAC;gBACf,MAAM,MAAM,CAAC,KAAK,CAAC;YACrB,CAAC;QACH,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,oEAAoE;YACpE,IAAI,OAAO,IAAI,OAAO,KAAK,UAAU,EAAE,CAAC;gBACtC,MAAM,KAAK,CAAC;YACd,CAAC;YACD,MAAM,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YACvE,MAAM,CAAC,KAAK,CAAC,IAAI,aAAa,oBAAoB,OAAO,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;YAC3E,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC;YAC/B,MAAM,CAAC,KAAK,CACV,IAAI,aAAa,iBAAiB,OAAO,kBAAkB,OAAO,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,GAAG,CAC5F,CAAC;YACF,MAAM,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC;QAC/D,CAAC;gBAAS,CAAC;YACT,uEAAuE;YACvE,uEAAuE;YACvE,uEAAuE;YACvE,8DAA8D;YAC9D,wEAAwE;YACxE,yEAAyE;YACzE,qBAAqB;YACrB,IAAI,CAAC,SAAS,EAAE,CAAC;gBACf,UAAU,CAAC,KAAK,EAAE,CAAC;gBACnB,KAAK,QAAQ,CAAC,MAAM,EAAE,CAAC,SAAS,CAAC,EAAE,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;YAC5D,CAAC;QACH,CAAC;IACH,CAAC;AACH,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@falai/agent",
-  "version": "2.4.3",
+  "version": "2.5.0",
   "description": "Conversational state engine for TypeScript where the AI understands, but the code is in control",
   "type": "module",
   "main": "./dist/cjs/index.js",

package/src/core/ToolLoopExecutor.ts

@@ -26,7 +26,7 @@ import type { Flow } from "./Flow";
 import type { Step } from "./Step";
 import type { ToolManager } from "./ToolManager";
 import { ResponseGenerationError } from "./ResponseGenerationError";
-import { historyToEvents, logger, serializeToolResult } from "../utils";
+import { historyToEvents, logger, serializeToolResult, assistantMessage, toolMessage } from "../utils";
 
 export class ToolLoopExecutor<TContext = unknown, TData = unknown> {
     constructor(
@@ -72,7 +72,10 @@ export class ToolLoopExecutor<TContext = unknown, TData = unknown> {
     }> {
         try {
             const { context, history, selectedFlow, responsePrompt, availableTools, responseSchema, signal } = params;
-            let { toolCalls, session } = params;
+            // The follow-up loop (which reassigned toolCalls) now lives in
+            // runFollowUpLoop; here toolCalls is only read.
+            const { toolCalls } = params;
+            let { session } = params;
 
             // Convert HistoryItem[] to Event[] for internal processing
             const historyEvents = historyToEvents(history);
@@ -146,6 +149,66 @@ export class ToolLoopExecutor<TContext = unknown, TData = unknown> {
                 }
             }
 
+            // Hand off to the multi-round follow-up loop shared with the
+            // streaming path. The initial batch above already executed
+            // sequentially and populated the result maps.
+            return await this.runFollowUpLoop({
+                toolCalls,
+                session,
+                toolResultsMap,
+                toolArgsMap,
+                context,
+                history,
+                selectedFlow,
+                responsePrompt,
+                availableTools,
+                responseSchema,
+                signal,
+            });
+        } catch (error) {
+            throw ResponseGenerationError.fromError(error, 'tool_execution', params, {
+                toolCallsCount: params.toolCalls?.length || 0,
+                availableToolsCount: params.availableTools.length
+            });
+        }
+    }
+
+    /**
+     * The multi-round follow-up loop shared by the non-streaming (`runLoop`) and
+     * streaming (`runStreamingBatch`) tool paths: re-prompt the model with the
+     * tool results (tools available on the first round so it can chain further
+     * calls), execute any further tool calls, repeat up to `maxToolLoops`, then
+     * force a result-aware closing message if the model never produced one.
+     * Callers run the *initial* batch — sequentially for `runLoop`, concurrently
+     * (with progress) for `runStreamingBatch` — and pass the populated result
+     * maps; from here both paths behave identically.
+     */
+    private async runFollowUpLoop(params: {
+        toolCalls?: Array<{ toolName: string; arguments: Record<string, unknown> }>;
+        session: SessionState<TData>;
+        toolResultsMap: Map<string, string>;
+        toolArgsMap: Map<string, Record<string, unknown>>;
+        context: TContext;
+        history: HistoryItem[];
+        selectedFlow?: Flow<TContext, TData>;
+        responsePrompt: string;
+        availableTools: Array<{
+            id: string;
+            name: string;
+            description?: string;
+            parameters?: unknown;
+        }>;
+        responseSchema?: Record<string, unknown>;
+        signal?: AbortSignal;
+    }): Promise<{
+        session: SessionState<TData>;
+        finalToolCalls?: Array<{ toolName: string; arguments: Record<string, unknown> }>;
+        finalMessage?: string;
+        structured?: AgentStructuredResponse;
+    }> {
+        const { context, history, selectedFlow, responsePrompt, availableTools, responseSchema, signal, toolResultsMap, toolArgsMap } = params;
+        let { toolCalls, session } = params;
+        try {
             // TOOL LOOP: Allow AI to make follow-up tool calls after initial tool execution
             // `??` so an explicit `maxToolLoops: 0` is honored instead of being
             // clobbered to the default by a falsy-zero check.
@@ -164,24 +227,18 @@ export class ToolLoopExecutor<TContext = unknown, TData = unknown> {
                 for (const toolCall of toolCalls || []) {
                     const tool = this.findAvailableTool(toolCall.toolName, selectedFlow);
                     if (tool) {
-                        // Create HistoryItem format for tool results
-                        // Add assistant message with tool_calls
-                        toolResultHistoryItems.push({
-                            role: "assistant" as const,
-                            content: null,
-                            tool_calls: [{
-                                id: toolCall.toolName,
-                                name: toolCall.toolName,
-                                arguments: toolCall.arguments,
-                            }],
-                        });
-                        // Add tool result
-                        toolResultHistoryItems.push({
-                            role: "tool" as const,
-                            tool_call_id: toolCall.toolName,
-                            name: toolCall.toolName,
-                            content: toolResultsMap.get(toolCall.toolName) || "Tool executed successfully",
-                        });
+                        // assistant tool_call + tool result pair, via the shared
+                        // history factories (same as forceFinalTextFromTools).
+                        toolResultHistoryItems.push(
+                            assistantMessage(null, [
+                                { id: toolCall.toolName, name: toolCall.toolName, arguments: toolCall.arguments },
+                            ]),
+                            toolMessage(
+                                toolCall.toolName,
+                                toolCall.toolName,
+                                toolResultsMap.get(toolCall.toolName) || "Tool executed successfully",
+                            ),
+                        );
                     }
                 }
 
@@ -364,21 +421,12 @@ export class ToolLoopExecutor<TContext = unknown, TData = unknown> {
         // call can see what the tools returned.
         const finalToolResultHistoryItems: HistoryItem[] = [];
         for (const [toolName, toolResult] of toolResultsMap) {
-            finalToolResultHistoryItems.push({
-                role: "assistant" as const,
-                content: null,
-                tool_calls: [{
-                    id: toolName,
-                    name: toolName,
-                    arguments: toolArgsMap.get(toolName) || {},
-                }],
-            });
-            finalToolResultHistoryItems.push({
-                role: "tool" as const,
-                tool_call_id: toolName,
-                name: toolName,
-                content: toolResult,
-            });
+            finalToolResultHistoryItems.push(
+                assistantMessage(null, [
+                    { id: toolName, name: toolName, arguments: toolArgsMap.get(toolName) || {} },
+                ]),
+                toolMessage(toolName, toolName, toolResult),
+            );
         }
 
         const finalHistory = [...history, ...finalToolResultHistoryItems];
@@ -522,25 +570,33 @@ export class ToolLoopExecutor<TContext = unknown, TData = unknown> {
 
             logger.debug(`[ToolLoopExecutor] Concurrent tool execution completed for ${toolCallRequests.length} tools`);
 
-            // Tools ran but streaming produced no post-tool message — force one
-            // from the results, mirroring runLoop. (The streaming chunk only ever
-            // carries the pre-tool preamble, never a result-aware response.)
-            if (toolResultsMap.size > 0) {
-                const forced = await this.forceFinalTextFromTools({
-                    history,
-                    toolResultsMap,
-                    toolArgsMap,
-                    responsePrompt,
-                    responseSchema,
-                    context,
-                    signal,
-                });
-                finalMessage = forced.finalMessage;
-                structured = forced.structured;
-            }
+            // Multi-round follow-up shared with the non-streaming path: re-prompt
+            // with the tool results so the model can chain further tool calls,
+            // then produce result-aware closing text (forced if it never does).
+            // Previously a single forced-text call — streaming now loops to parity
+            // with runLoop, so a streamed turn can chain tools across rounds.
+            const followUp = await this.runFollowUpLoop({
+                toolCalls,
+                session,
+                toolResultsMap,
+                toolArgsMap,
+                context,
+                history,
+                selectedFlow,
+                responsePrompt,
+                availableTools,
+                responseSchema,
+                signal,
+            });
+            session = followUp.session;
+            toolCalls = followUp.finalToolCalls;
+            finalMessage = followUp.finalMessage;
+            structured = followUp.structured;
         } catch (error) {
             logger.error(`[ToolLoopExecutor] Concurrent tool execution failed, falling back to sequential:`, error);
-            // Fall back to the unified tool loop on failure
+            // Fall back to the unified tool loop on failure. runLoop re-executes
+            // the tools from scratch, so any partial results collected above are
+            // intentionally discarded (it builds and forces its own).
             const toolResult = await this.runLoop({
                 toolCalls, context, session, history, selectedFlow,
                 responsePrompt, availableTools, responseSchema, signal,

package/src/index.ts

@@ -32,6 +32,9 @@ export { DeepSeekProvider } from "./providers/DeepSeekProvider";
 export type { DeepSeekProviderOptions } from "./providers/DeepSeekProvider";
 // Base class for building OpenAI-compatible providers (Groq, Together, etc.)
 export { OpenAICompatibleProvider } from "./providers/OpenAICompatibleProvider";
+export type { StructuredOutputMode } from "./providers/OpenAICompatibleProvider";
+export { createOpenAICompatibleProvider } from "./providers/GenericOpenAICompatibleProvider";
+export type { OpenAICompatibleOptions } from "./providers/GenericOpenAICompatibleProvider";
 
 // Adapters
 export { PrismaAdapter } from "./adapters/PrismaAdapter";

package/src/providers/AnthropicProvider.ts

@@ -17,7 +17,7 @@ import type {
 } from "../types";
 import type { ProviderCapabilities } from "../types/ai";
 import type { HistoryItem } from "../types/history";
-import { withTimeoutAndRetry, withStreamRetry, logger } from "../utils";
+import { withTimeoutAndRetry, withStreamRetry, resolveRetryConfig, logger, assertUsableCompletion, combineAbortSignals } from "../utils";
 import {
   classifyProviderError,
   getErrorMessage,
@@ -26,11 +26,6 @@ import {
   type ErrorClassificationOptions,
 } from "./errorClassification";
 
-const DEFAULT_RETRY_CONFIG = {
-  timeout: 60000,
-  retries: 3,
-};
-
 /**
  * Configuration options for Anthropic provider
  * Uses types from @anthropic-ai/sdk package
@@ -105,14 +100,7 @@ export class AnthropicProvider implements AiProvider {
     this.primaryModel = model;
     this.backupModels = backupModels;
     this.config = config;
-    this.retryConfig = {
-      // `||` is intentional: a 0ms timeout is degenerate (aborts every call
-      // immediately), so fall back to the default.
-      timeout: retryConfig?.timeout || DEFAULT_RETRY_CONFIG.timeout,
-      // `??` so an explicit `retries: 0` (disable retries) is honored instead of
-      // being clobbered to the default by a falsy-zero check.
-      retries: retryConfig?.retries ?? DEFAULT_RETRY_CONFIG.retries,
-    };
+    this.retryConfig = resolveRetryConfig(retryConfig);
   }
 
   /**
@@ -268,7 +256,7 @@ export class AnthropicProvider implements AiProvider {
     model: string,
     input: GenerateMessageInput<TContext>
   ): Promise<GenerateMessageOutput<TStructured>> {
-    const operation = async (): Promise<GenerateMessageOutput> => {
+    const operation = async (signal: AbortSignal): Promise<GenerateMessageOutput> => {
       // Anthropic requires max_tokens to be specified
       const maxTokens = input.parameters?.maxOutputTokens || 4096;
 
@@ -332,7 +320,9 @@ export class AnthropicProvider implements AiProvider {
         }
       }
 
-      const response = await this.client.messages.create(params);
+      const response = await this.client.messages.create(params, {
+        signal: combineAbortSignals(input.signal, signal),
+      });
 
       // Extract text and tool calls from response
       const textContent = response.content.find(
@@ -381,16 +371,10 @@ export class AnthropicProvider implements AiProvider {
         } as AgentStructuredResponse;
       }
 
-      // A parsed-but-blank structured message with no tool calls is just as
-      // empty as no text at all — throw so withTimeoutAndRetry retries instead
-      // of returning {"message":""}.
-      if (
-        toolCalls.length === 0 &&
-        typeof structured?.message === "string" &&
-        !structured.message.trim()
-      ) {
-        throw new Error("No response from Anthropic");
-      }
+      // A parsed-but-blank message with no tool calls is as empty as no text;
+      // the shared guard throws so withTimeoutAndRetry retries instead of
+      // returning {"message":""}.
+      assertUsableCompletion(structured, message, toolCalls.length, "Anthropic");
 
       return {
         message,
@@ -423,8 +407,8 @@ export class AnthropicProvider implements AiProvider {
     // Try primary model first
     try {
       yield* withStreamRetry(
-        () => this.generateStreamWithModel<TContext, TStructured>(this.primaryModel, input),
-        { maxRetries: this.retryConfig.retries, operationName: `Anthropic ${this.primaryModel} stream` }
+        (signal) => this.generateStreamWithModel<TContext, TStructured>(this.primaryModel, input, signal),
+        { maxRetries: this.retryConfig.retries, firstChunkTimeoutMs: this.retryConfig.timeout, operationName: `Anthropic ${this.primaryModel} stream` }
       );
     } catch (primaryError: unknown) {
       const primaryErrMsg = getErrorMessage(primaryError);
@@ -449,8 +433,8 @@ export class AnthropicProvider implements AiProvider {
 
         try {
           yield* withStreamRetry(
-            () => this.generateStreamWithModel<TContext, TStructured>(backupModel, input),
-            { maxRetries: this.retryConfig.retries, operationName: `Anthropic ${backupModel} stream` }
+            (signal) => this.generateStreamWithModel<TContext, TStructured>(backupModel, input, signal),
+            { maxRetries: this.retryConfig.retries, firstChunkTimeoutMs: this.retryConfig.timeout, operationName: `Anthropic ${backupModel} stream` }
           );
           logger.debug(`[ANTHROPIC] Backup model ${backupModel} succeeded`);
           return;
@@ -486,7 +470,8 @@ export class AnthropicProvider implements AiProvider {
     TStructured = AgentStructuredResponse
   >(
     model: string,
-    input: GenerateMessageInput<TContext>
+    input: GenerateMessageInput<TContext>,
+    attemptSignal?: AbortSignal
   ): AsyncGenerator<GenerateMessageStreamChunk<TStructured>> {
     // Anthropic requires max_tokens to be specified
     const maxTokens = input.parameters?.maxOutputTokens || 4096;
@@ -551,7 +536,9 @@ export class AnthropicProvider implements AiProvider {
       }
     }
 
-    const stream = this.client.messages.stream(params);
+    const stream = this.client.messages.stream(params, {
+      signal: combineAbortSignals(input.signal, attemptSignal),
+    });
 
     let accumulated = "";
     let currentModel = model;
@@ -612,17 +599,9 @@ export class AnthropicProvider implements AiProvider {
       } as AgentStructuredResponse;
     }
 
-    // Empty-completion guard — mirror of the non-streaming path. A blank
-    // effective message (structured message under a schema, else accumulated
-    // text) with no tool calls means the model produced nothing usable; throw
-    // so withStreamRetry/generateStreamWithBackup retry instead of silently
-    // emitting an empty message.
-    const messageText = (
-      typeof structured?.message === "string" ? structured.message : accumulated
-    ).trim();
-    if (!messageText && toolCalls.length === 0) {
-      throw new Error("No response from Anthropic");
-    }
+    // Empty-completion guard — same definition as the non-streaming path, so the
+    // stream retries / falls back to backup instead of emitting an empty message.
+    assertUsableCompletion(structured, accumulated, toolCalls.length, "Anthropic");
 
     // Yield final chunk
     yield {