@tensorcost/sdk 0.3.0 → 0.5.0

package/dist/proxy-client.js

@@ -0,0 +1,259 @@
+/**
+ * Applied-mode proxy HTTP client.
+ *
+ * This is the single place that combines retries, timeouts, typed errors,
+ * telemetry hooks, and the fail-open circuit breaker. The provider wrappers
+ * (openai.ts, anthropic.ts) call `proxyRequest()` instead of hitting the
+ * proxy directly.
+ *
+ * The contract:
+ *   - Returns the parsed response body on success.
+ *   - On all non-success outcomes, throws a subclass of TensorCostError.
+ *   - Never swallows errors — the provider wrapper's fail-open logic
+ *     (fall back to direct call on proxy failure) lives in the provider
+ *     wrapper, not here.
+ *
+ * Timeout handling:
+ *   - Non-streaming: AbortController fires after `timeoutMs` total.
+ *   - Streaming is not handled in v1 (no stream: true support yet).
+ *
+ * Retry logic:
+ *   - Retries on 5xx + network errors + 429.
+ *   - Respects Retry-After on 429.
+ *   - Stops at maxAttempts.
+ *   - Does not retry on 4xx (except 429).
+ *
+ * Circuit breaker:
+ *   - Maintained on the CircuitBreaker instance passed in.
+ *   - When open, throws TensorCostProxyError immediately with a clear
+ *     message so the provider wrapper can activate the direct-provider
+ *     fallback.
+ */
+import { TensorCostError, TensorCostNetworkError, TensorCostTimeoutError, TensorCostProxyError, TensorCostQuotaError, TensorCostProviderError, } from "./errors.js";
+import { DEFAULT_RETRY_CONFIG, computeDelay, retryAfterMs, sleep, } from "./retry.js";
+import { emitEvent } from "./telemetry.js";
+const PROXY_CHAT_PATH = "/api/inference-proxy/v1/chat/completions";
+/**
+ * Execute a single proxied request with retries, timeouts, telemetry,
+ * and circuit-breaker integration.
+ *
+ * Throws a `TensorCostError` subclass on any failure.
+ * Returns the parsed JSON body on success.
+ */
+export async function proxyRequest(opts) {
+    const retryConfig = opts.retry ?? DEFAULT_RETRY_CONFIG;
+    const timeoutMs = opts.timeoutMs ?? 60_000;
+    const fetchImpl = opts.fetchImpl ?? globalThis.fetch.bind(globalThis);
+    const callStart = Date.now();
+    // Check circuit breaker — if open, throw immediately so the caller can
+    // activate the direct-provider fallback path.
+    if (opts.circuit.shouldBypass()) {
+        throw new TensorCostProxyError("tensorcost: proxy circuit open — routing directly to provider", { attempt: 1 });
+    }
+    const url = `${opts.proxyUrl.replace(/\/+$/, "")}${PROXY_CHAT_PATH}`;
+    const headers = {
+        Authorization: `Bearer ${opts.bearerToken}`,
+        "Content-Type": "application/json",
+        "x-tc-provider-url": opts.providerUrl,
+        "x-tc-provider-auth": opts.providerAuth,
+        "x-tc-correlation-id": opts.correlationId,
+    };
+    if (opts.environment) {
+        headers["x-tc-environment"] = opts.environment;
+    }
+    let lastError = null;
+    for (let attempt = 0; attempt < retryConfig.maxAttempts; attempt++) {
+        const elapsedMs = Date.now() - callStart;
+        emitEvent(opts.onLifecycleEvent, {
+            kind: "before_request",
+            provider: opts.provider,
+            model: opts.model,
+            operation: opts.operation,
+            attemptNumber: attempt + 1,
+            elapsedMs,
+        });
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), timeoutMs);
+        let resp;
+        try {
+            resp = await fetchImpl(url, {
+                method: "POST",
+                headers,
+                body: JSON.stringify(opts.body),
+                signal: controller.signal,
+            });
+        }
+        catch (err) {
+            clearTimeout(timer);
+            // AbortController fired → timeout.
+            if (err instanceof Error &&
+                (err.name === "AbortError" || err.name === "TimeoutError")) {
+                const tcErr = new TensorCostTimeoutError(`tensorcost: proxy request timed out after ${timeoutMs}ms`, "total", { cause: err, attempt: attempt + 1 });
+                lastError = tcErr;
+                opts.circuit.recordFailure();
+                // Timeouts are retriable.
+                if (attempt + 1 < retryConfig.maxAttempts) {
+                    const delay = computeDelay(attempt, retryConfig);
+                    emitEvent(opts.onLifecycleEvent, {
+                        kind: "on_retry",
+                        provider: opts.provider,
+                        model: opts.model,
+                        operation: opts.operation,
+                        attemptNumber: attempt + 1,
+                        elapsedMs: Date.now() - callStart,
+                        reason: "network_error",
+                        status: null,
+                        delayMs: delay,
+                    });
+                    await sleep(delay);
+                    continue;
+                }
+                emitEvent(opts.onLifecycleEvent, {
+                    kind: "on_error",
+                    provider: opts.provider,
+                    model: opts.model,
+                    operation: opts.operation,
+                    attemptNumber: attempt + 1,
+                    elapsedMs: Date.now() - callStart,
+                    errorClass: tcErr.constructor.name,
+                    status: null,
+                });
+                throw tcErr;
+            }
+            // Network error (ECONNREFUSED, DNS failure, etc.)
+            const tcErr = new TensorCostNetworkError(`tensorcost: network error reaching proxy: ${err.message}`, { cause: err, attempt: attempt + 1 });
+            lastError = tcErr;
+            opts.circuit.recordFailure();
+            if (attempt + 1 < retryConfig.maxAttempts) {
+                const delay = computeDelay(attempt, retryConfig);
+                emitEvent(opts.onLifecycleEvent, {
+                    kind: "on_retry",
+                    provider: opts.provider,
+                    model: opts.model,
+                    operation: opts.operation,
+                    attemptNumber: attempt + 1,
+                    elapsedMs: Date.now() - callStart,
+                    reason: "network_error",
+                    status: null,
+                    delayMs: delay,
+                });
+                await sleep(delay);
+                continue;
+            }
+            emitEvent(opts.onLifecycleEvent, {
+                kind: "on_error",
+                provider: opts.provider,
+                model: opts.model,
+                operation: opts.operation,
+                attemptNumber: attempt + 1,
+                elapsedMs: Date.now() - callStart,
+                errorClass: tcErr.constructor.name,
+                status: null,
+            });
+            throw tcErr;
+        }
+        finally {
+            clearTimeout(timer);
+        }
+        const requestId = resp.headers.get("x-tc-request-id");
+        // 2xx — success.
+        if (resp.ok) {
+            opts.circuit.recordSuccess();
+            emitEvent(opts.onLifecycleEvent, {
+                kind: "after_response",
+                provider: opts.provider,
+                model: opts.model,
+                operation: opts.operation,
+                attemptNumber: attempt + 1,
+                elapsedMs: Date.now() - callStart,
+                status: resp.status,
+                requestId,
+            });
+            return resp.json();
+        }
+        // 429 — quota error, may have Retry-After.
+        if (resp.status === 429) {
+            const raMs = retryAfterMs(resp);
+            const tcErr = new TensorCostQuotaError(`tensorcost: proxy returned 429 Too Many Requests`, raMs ?? null, { status: 429, requestId, attempt: attempt + 1 });
+            lastError = tcErr;
+            // Don't count quota errors against the circuit (the proxy is healthy).
+            if (attempt + 1 < retryConfig.maxAttempts) {
+                const delay = computeDelay(attempt, retryConfig, raMs);
+                emitEvent(opts.onLifecycleEvent, {
+                    kind: "on_retry",
+                    provider: opts.provider,
+                    model: opts.model,
+                    operation: opts.operation,
+                    attemptNumber: attempt + 1,
+                    elapsedMs: Date.now() - callStart,
+                    reason: "quota_429",
+                    status: 429,
+                    delayMs: delay,
+                });
+                await sleep(delay);
+                continue;
+            }
+            emitEvent(opts.onLifecycleEvent, {
+                kind: "on_error",
+                provider: opts.provider,
+                model: opts.model,
+                operation: opts.operation,
+                attemptNumber: attempt + 1,
+                elapsedMs: Date.now() - callStart,
+                errorClass: tcErr.constructor.name,
+                status: 429,
+            });
+            throw tcErr;
+        }
+        // 5xx — proxy infrastructure error.
+        if (resp.status >= 500) {
+            const tcErr = new TensorCostProxyError(`tensorcost: proxy returned ${resp.status}`, { status: resp.status, requestId, attempt: attempt + 1 });
+            lastError = tcErr;
+            opts.circuit.recordFailure();
+            if (attempt + 1 < retryConfig.maxAttempts) {
+                const delay = computeDelay(attempt, retryConfig);
+                emitEvent(opts.onLifecycleEvent, {
+                    kind: "on_retry",
+                    provider: opts.provider,
+                    model: opts.model,
+                    operation: opts.operation,
+                    attemptNumber: attempt + 1,
+                    elapsedMs: Date.now() - callStart,
+                    reason: "5xx",
+                    status: resp.status,
+                    delayMs: delay,
+                });
+                await sleep(delay);
+                continue;
+            }
+            emitEvent(opts.onLifecycleEvent, {
+                kind: "on_error",
+                provider: opts.provider,
+                model: opts.model,
+                operation: opts.operation,
+                attemptNumber: attempt + 1,
+                elapsedMs: Date.now() - callStart,
+                errorClass: tcErr.constructor.name,
+                status: resp.status,
+            });
+            throw tcErr;
+        }
+        // 4xx (not 429) — provider error forwarded by the proxy. Non-retriable.
+        const tcErr = new TensorCostProviderError(`tensorcost: upstream provider returned ${resp.status}`, { status: resp.status, requestId, attempt: attempt + 1 });
+        emitEvent(opts.onLifecycleEvent, {
+            kind: "on_error",
+            provider: opts.provider,
+            model: opts.model,
+            operation: opts.operation,
+            attemptNumber: attempt + 1,
+            elapsedMs: Date.now() - callStart,
+            errorClass: tcErr.constructor.name,
+            status: resp.status,
+        });
+        throw tcErr;
+    }
+    // Should be unreachable — every loop arm either returns or throws.
+    throw lastError ?? new TensorCostProxyError("tensorcost: max attempts reached");
+}
+export { TensorCostError };
+//# sourceMappingURL=proxy-client.js.map

package/dist/proxy-client.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"proxy-client.js","sourceRoot":"","sources":["../src/proxy-client.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAEH,OAAO,EACL,eAAe,EACf,sBAAsB,EACtB,sBAAsB,EACtB,oBAAoB,EACpB,oBAAoB,EACpB,uBAAuB,GACxB,MAAM,aAAa,CAAC;AACrB,OAAO,EACL,oBAAoB,EACpB,YAAY,EAEZ,YAAY,EACZ,KAAK,GACN,MAAM,YAAY,CAAC;AAGpB,OAAO,EAAE,SAAS,EAAE,MAAM,gBAAgB,CAAC;AAyB3C,MAAM,eAAe,GAAG,0CAA0C,CAAC;AAEnE;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,IAAyB;IAEzB,MAAM,WAAW,GAAG,IAAI,CAAC,KAAK,IAAI,oBAAoB,CAAC;IACvD,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,MAAM,CAAC;IAC3C,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;IACtE,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAE7B,uEAAuE;IACvE,8CAA8C;IAC9C,IAAI,IAAI,CAAC,OAAO,CAAC,YAAY,EAAE,EAAE,CAAC;QAChC,MAAM,IAAI,oBAAoB,CAC5B,+DAA+D,EAC/D,EAAE,OAAO,EAAE,CAAC,EAAE,CACf,CAAC;IACJ,CAAC;IAED,MAAM,GAAG,GAAG,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,GAAG,eAAe,EAAE,CAAC;IACrE,MAAM,OAAO,GAA2B;QACtC,aAAa,EAAE,UAAU,IAAI,CAAC,WAAW,EAAE;QAC3C,cAAc,EAAE,kBAAkB;QAClC,mBAAmB,EAAE,IAAI,CAAC,WAAW;QACrC,oBAAoB,EAAE,IAAI,CAAC,YAAY;QACvC,qBAAqB,EAAE,IAAI,CAAC,aAAa;KAC1C,CAAC;IACF,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;QACrB,OAAO,CAAC,kBAAkB,CAAC,GAAG,IAAI,CAAC,WAAW,CAAC;IACjD,CAAC;IAED,IAAI,SAAS,GAA2B,IAAI,CAAC;IAE7C,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,WAAW,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC;QACnE,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;QAEzC,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;YAC/B,IAAI,EAAE,gBAAgB;YACtB,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,aAAa,EAAE,OAAO,GAAG,CAAC;YAC1B,SAAS;SACV,CAAC,CAAC;QAEH,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;QAE9D,IAAI,IAAc,CAAC;QACnB,IAAI,CAAC;YACH,IAAI,GAAG,MAAM,SAAS,CAAC,GAAG,EAAE;gBAC1B,MAAM,EAAE,MAAM;gBACd,OAAO;gBACP,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC;gBAC/B,MAAM,EAAE,UAAU,CAAC,MAAM;aAC1B,CAAC,CAAC;QACL,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,YAAY,CAAC,KAAK,CAAC,CAAC;YAEpB,mCAAmC;YACnC,IACE,GAAG,YAAY,KAAK;gBACpB,CAAC,GAAG,CAAC,IAAI,KAAK,YAAY,IAAI,GAAG,CAAC,IAAI,KAAK,cAAc,CAAC,EAC1D,CAAC;gBACD,MAAM,KAAK,GAAG,IAAI,sBAAsB,CACtC,6CAA6C,SAAS,IAAI,EAC1D,OAAO,EACP,EAAE,KAAK,EAAE,GAAG,EAAE,OAAO,EAAE,OAAO,GAAG,CAAC,EAAE,CACrC,CAAC;gBACF,SAAS,GAAG,KAAK,CAAC;gBAClB,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,CAAC;gBAE7B,0BAA0B;gBAC1B,IAAI,OAAO,GAAG,CAAC,GAAG,WAAW,CAAC,WAAW,EAAE,CAAC;oBAC1C,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;oBACjD,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;wBAC/B,IAAI,EAAE,UAAU;wBAChB,QAAQ,EAAE,IAAI,CAAC,QAAQ;wBACvB,KAAK,EAAE,IAAI,CAAC,KAAK;wBACjB,SAAS,EAAE,IAAI,CAAC,SAAS;wBACzB,aAAa,EAAE,OAAO,GAAG,CAAC;wBAC1B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBACjC,MAAM,EAAE,eAAe;wBACvB,MAAM,EAAE,IAAI;wBACZ,OAAO,EAAE,KAAK;qBACf,CAAC,CAAC;oBACH,MAAM,KAAK,CAAC,KAAK,CAAC,CAAC;oBACnB,SAAS;gBACX,CAAC;gBAED,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;oBAC/B,IAAI,EAAE,UAAU;oBAChB,QAAQ,EAAE,IAAI,CAAC,QAAQ;oBACvB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,SAAS,EAAE,IAAI,CAAC,SAAS;oBACzB,aAAa,EAAE,OAAO,GAAG,CAAC;oBAC1B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBACjC,UAAU,EAAE,KAAK,CAAC,WAAW,CAAC,IAAI;oBAClC,MAAM,EAAE,IAAI;iBACb,CAAC,CAAC;gBACH,MAAM,KAAK,CAAC;YACd,CAAC;YAED,kDAAkD;YAClD,MAAM,KAAK,GAAG,IAAI,sBAAsB,CACtC,6CAA8C,GAAa,CAAC,OAAO,EAAE,EACrE,EAAE,KAAK,EAAE,GAAY,EAAE,OAAO,EAAE,OAAO,GAAG,CAAC,EAAE,CAC9C,CAAC;YACF,SAAS,GAAG,KAAK,CAAC;YAClB,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,CAAC;YAE7B,IAAI,OAAO,GAAG,CAAC,GAAG,WAAW,CAAC,WAAW,EAAE,CAAC;gBAC1C,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;gBACjD,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;oBAC/B,IAAI,EAAE,UAAU;oBAChB,QAAQ,EAAE,IAAI,CAAC,QAAQ;oBACvB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,SAAS,EAAE,IAAI,CAAC,SAAS;oBACzB,aAAa,EAAE,OAAO,GAAG,CAAC;oBAC1B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBACjC,MAAM,EAAE,eAAe;oBACvB,MAAM,EAAE,IAAI;oBACZ,OAAO,EAAE,KAAK;iBACf,CAAC,CAAC;gBACH,MAAM,KAAK,CAAC,KAAK,CAAC,CAAC;gBACnB,SAAS;YACX,CAAC;YAED,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;gBAC/B,IAAI,EAAE,UAAU;gBAChB,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,aAAa,EAAE,OAAO,GAAG,CAAC;gBAC1B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBACjC,UAAU,EAAE,KAAK,CAAC,WAAW,CAAC,IAAI;gBAClC,MAAM,EAAE,IAAI;aACb,CAAC,CAAC;YACH,MAAM,KAAK,CAAC;QACd,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,KAAK,CAAC,CAAC;QACtB,CAAC;QAED,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAC;QAEtD,iBAAiB;QACjB,IAAI,IAAI,CAAC,EAAE,EAAE,CAAC;YACZ,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,CAAC;YAC7B,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;gBAC/B,IAAI,EAAE,gBAAgB;gBACtB,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,aAAa,EAAE,OAAO,GAAG,CAAC;gBAC1B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBACjC,MAAM,EAAE,IAAI,CAAC,MAAM;gBACnB,SAAS;aACV,CAAC,CAAC;YACH,OAAO,IAAI,CAAC,IAAI,EAAsB,CAAC;QACzC,CAAC;QAED,2CAA2C;QAC3C,IAAI,IAAI,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;YACxB,MAAM,IAAI,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;YAChC,MAAM,KAAK,GAAG,IAAI,oBAAoB,CACpC,kDAAkD,EAClD,IAAI,IAAI,IAAI,EACZ,EAAE,MAAM,EAAE,GAAG,EAAE,SAAS,EAAE,OAAO,EAAE,OAAO,GAAG,CAAC,EAAE,CACjD,CAAC;YACF,SAAS,GAAG,KAAK,CAAC;YAClB,uEAAuE;YAEvE,IAAI,OAAO,GAAG,CAAC,GAAG,WAAW,CAAC,WAAW,EAAE,CAAC;gBAC1C,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,EAAE,WAAW,EAAE,IAAI,CAAC,CAAC;gBACvD,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;oBAC/B,IAAI,EAAE,UAAU;oBAChB,QAAQ,EAAE,IAAI,CAAC,QAAQ;oBACvB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,SAAS,EAAE,IAAI,CAAC,SAAS;oBACzB,aAAa,EAAE,OAAO,GAAG,CAAC;oBAC1B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBACjC,MAAM,EAAE,WAAW;oBACnB,MAAM,EAAE,GAAG;oBACX,OAAO,EAAE,KAAK;iBACf,CAAC,CAAC;gBACH,MAAM,KAAK,CAAC,KAAK,CAAC,CAAC;gBACnB,SAAS;YACX,CAAC;YAED,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;gBAC/B,IAAI,EAAE,UAAU;gBAChB,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,aAAa,EAAE,OAAO,GAAG,CAAC;gBAC1B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBACjC,UAAU,EAAE,KAAK,CAAC,WAAW,CAAC,IAAI;gBAClC,MAAM,EAAE,GAAG;aACZ,CAAC,CAAC;YACH,MAAM,KAAK,CAAC;QACd,CAAC;QAED,oCAAoC;QACpC,IAAI,IAAI,CAAC,MAAM,IAAI,GAAG,EAAE,CAAC;YACvB,MAAM,KAAK,GAAG,IAAI,oBAAoB,CACpC,8BAA8B,IAAI,CAAC,MAAM,EAAE,EAC3C,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,SAAS,EAAE,OAAO,EAAE,OAAO,GAAG,CAAC,EAAE,CACzD,CAAC;YACF,SAAS,GAAG,KAAK,CAAC;YAClB,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,CAAC;YAE7B,IAAI,OAAO,GAAG,CAAC,GAAG,WAAW,CAAC,WAAW,EAAE,CAAC;gBAC1C,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;gBACjD,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;oBAC/B,IAAI,EAAE,UAAU;oBAChB,QAAQ,EAAE,IAAI,CAAC,QAAQ;oBACvB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,SAAS,EAAE,IAAI,CAAC,SAAS;oBACzB,aAAa,EAAE,OAAO,GAAG,CAAC;oBAC1B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBACjC,MAAM,EAAE,KAAK;oBACb,MAAM,EAAE,IAAI,CAAC,MAAM;oBACnB,OAAO,EAAE,KAAK;iBACf,CAAC,CAAC;gBACH,MAAM,KAAK,CAAC,KAAK,CAAC,CAAC;gBACnB,SAAS;YACX,CAAC;YAED,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;gBAC/B,IAAI,EAAE,UAAU;gBAChB,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,aAAa,EAAE,OAAO,GAAG,CAAC;gBAC1B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBACjC,UAAU,EAAE,KAAK,CAAC,WAAW,CAAC,IAAI;gBAClC,MAAM,EAAE,IAAI,CAAC,MAAM;aACpB,CAAC,CAAC;YACH,MAAM,KAAK,CAAC;QACd,CAAC;QAED,wEAAwE;QACxE,MAAM,KAAK,GAAG,IAAI,uBAAuB,CACvC,0CAA0C,IAAI,CAAC,MAAM,EAAE,EACvD,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,SAAS,EAAE,OAAO,EAAE,OAAO,GAAG,CAAC,EAAE,CACzD,CAAC;QACF,SAAS,CAAC,IAAI,CAAC,gBAAgB,EAAE;YAC/B,IAAI,EAAE,UAAU;YAChB,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,aAAa,EAAE,OAAO,GAAG,CAAC;YAC1B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;YACjC,UAAU,EAAE,KAAK,CAAC,WAAW,CAAC,IAAI;YAClC,MAAM,EAAE,IAAI,CAAC,MAAM;SACpB,CAAC,CAAC;QACH,MAAM,KAAK,CAAC;IACd,CAAC;IAED,mEAAmE;IACnE,MAAM,SAAS,IAAI,IAAI,oBAAoB,CAAC,kCAAkC,CAAC,CAAC;AAClF,CAAC;AAED,OAAO,EAAE,eAAe,EAAE,CAAC"}

package/dist/retry.d.ts

@@ -0,0 +1,50 @@
+/**
+ * Retry utilities for the TensorCost applied-mode request path.
+ *
+ * Retries use jittered exponential backoff:
+ *
+ *   delay = min(maxDelayMs, baseDelayMs * 2^attempt) * (0.5 + Math.random() * 0.5)
+ *
+ * where `attempt` is 0-indexed. With defaults (baseDelayMs=500, maxDelayMs=30 000),
+ * the sequence of *maximum* delays is:
+ *   attempt 1: 500ms
+ *   attempt 2: 1 000ms
+ *   attempt 3: 2 000ms
+ *   attempt 4: 4 000ms
+ *   attempt 5: 8 000ms
+ * The jitter factor halves the minimum so actual delays are 50–100% of those
+ * maximums. This matches the posture used by the OpenAI Node SDK.
+ *
+ * Rules:
+ *   - Never retry 4xx responses (except 429).
+ *   - Always retry 5xx responses and network errors.
+ *   - On 429: honour the `Retry-After` response header when present
+ *     (overrides the backoff delay).
+ *   - Non-idempotent verbs (POST without `stream:false` is still idempotent
+ *     at the proxy level — the proxy is stateless) default to idempotent for
+ *     all chat/completions calls.
+ */
+export interface RetryConfig {
+    /** Maximum number of total attempts (1 = no retries). Default: 5. */
+    maxAttempts: number;
+    /** Base backoff interval in milliseconds. Default: 500. */
+    baseDelayMs: number;
+    /** Cap on any single backoff interval. Default: 30 000 (30s). */
+    maxDelayMs: number;
+}
+export declare const DEFAULT_RETRY_CONFIG: RetryConfig;
+/**
+ * Compute the delay before the next attempt.
+ *
+ * When `retryAfterMs` is provided (from a 429 Retry-After header) it
+ * overrides the exponential calculation but is still capped at maxDelayMs.
+ */
+export declare function computeDelay(attempt: number, // 0-indexed
+config: RetryConfig, retryAfterMs?: number): number;
+/** True for HTTP status codes the SDK should retry on. */
+export declare function isRetryableStatus(status: number): boolean;
+/** Extract Retry-After delay in milliseconds from a fetch Response. */
+export declare function retryAfterMs(resp: Response): number | undefined;
+/** Sleep for `ms` milliseconds (abortable). */
+export declare function sleep(ms: number, signal?: AbortSignal): Promise<void>;
+//# sourceMappingURL=retry.d.ts.map

package/dist/retry.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"retry.d.ts","sourceRoot":"","sources":["../src/retry.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAEH,MAAM,WAAW,WAAW;IAC1B,qEAAqE;IACrE,WAAW,EAAE,MAAM,CAAC;IACpB,2DAA2D;IAC3D,WAAW,EAAE,MAAM,CAAC;IACpB,iEAAiE;IACjE,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,eAAO,MAAM,oBAAoB,EAAE,WAIlC,CAAC;AAEF;;;;;GAKG;AACH,wBAAgB,YAAY,CAC1B,OAAO,EAAE,MAAM,EAAE,YAAY;AAC7B,MAAM,EAAE,WAAW,EACnB,YAAY,CAAC,EAAE,MAAM,GACpB,MAAM,CAQR;AAED,0DAA0D;AAC1D,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAEzD;AAED,uEAAuE;AACvE,wBAAgB,YAAY,CAAC,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,SAAS,CAS/D;AAED,+CAA+C;AAC/C,wBAAgB,KAAK,CAAC,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC,CAYrE"}

package/dist/retry.js

@@ -0,0 +1,80 @@
+/**
+ * Retry utilities for the TensorCost applied-mode request path.
+ *
+ * Retries use jittered exponential backoff:
+ *
+ *   delay = min(maxDelayMs, baseDelayMs * 2^attempt) * (0.5 + Math.random() * 0.5)
+ *
+ * where `attempt` is 0-indexed. With defaults (baseDelayMs=500, maxDelayMs=30 000),
+ * the sequence of *maximum* delays is:
+ *   attempt 1: 500ms
+ *   attempt 2: 1 000ms
+ *   attempt 3: 2 000ms
+ *   attempt 4: 4 000ms
+ *   attempt 5: 8 000ms
+ * The jitter factor halves the minimum so actual delays are 50–100% of those
+ * maximums. This matches the posture used by the OpenAI Node SDK.
+ *
+ * Rules:
+ *   - Never retry 4xx responses (except 429).
+ *   - Always retry 5xx responses and network errors.
+ *   - On 429: honour the `Retry-After` response header when present
+ *     (overrides the backoff delay).
+ *   - Non-idempotent verbs (POST without `stream:false` is still idempotent
+ *     at the proxy level — the proxy is stateless) default to idempotent for
+ *     all chat/completions calls.
+ */
+export const DEFAULT_RETRY_CONFIG = {
+    maxAttempts: 5,
+    baseDelayMs: 500,
+    maxDelayMs: 30_000,
+};
+/**
+ * Compute the delay before the next attempt.
+ *
+ * When `retryAfterMs` is provided (from a 429 Retry-After header) it
+ * overrides the exponential calculation but is still capped at maxDelayMs.
+ */
+export function computeDelay(attempt, // 0-indexed
+config, retryAfterMs) {
+    if (retryAfterMs !== null && retryAfterMs !== undefined && retryAfterMs >= 0) {
+        return Math.min(retryAfterMs, config.maxDelayMs);
+    }
+    const exponential = config.baseDelayMs * Math.pow(2, attempt);
+    const capped = Math.min(exponential, config.maxDelayMs);
+    // Jitter: uniform in [0.5 * capped, 1.0 * capped]
+    return capped * (0.5 + Math.random() * 0.5);
+}
+/** True for HTTP status codes the SDK should retry on. */
+export function isRetryableStatus(status) {
+    return status === 429 || status >= 500;
+}
+/** Extract Retry-After delay in milliseconds from a fetch Response. */
+export function retryAfterMs(resp) {
+    const header = resp.headers.get("retry-after");
+    if (!header)
+        return undefined;
+    // RFC 7231: numeric seconds OR HTTP-date
+    const secs = parseFloat(header);
+    if (!isNaN(secs))
+        return Math.ceil(secs * 1000);
+    const date = Date.parse(header);
+    if (!isNaN(date))
+        return Math.max(0, date - Date.now());
+    return undefined;
+}
+/** Sleep for `ms` milliseconds (abortable). */
+export function sleep(ms, signal) {
+    return new Promise((resolve, reject) => {
+        if (signal?.aborted) {
+            reject(new DOMException("Aborted", "AbortError"));
+            return;
+        }
+        const timer = setTimeout(resolve, ms);
+        signal?.addEventListener("abort", () => {
+            clearTimeout(timer);
+            reject(new DOMException("Aborted", "AbortError"));
+        });
+    });
+}
+//# sourceMappingURL=retry.js.map

package/dist/retry.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"retry.js","sourceRoot":"","sources":["../src/retry.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAWH,MAAM,CAAC,MAAM,oBAAoB,GAAgB;IAC/C,WAAW,EAAE,CAAC;IACd,WAAW,EAAE,GAAG;IAChB,UAAU,EAAE,MAAM;CACnB,CAAC;AAEF;;;;;GAKG;AACH,MAAM,UAAU,YAAY,CAC1B,OAAe,EAAE,YAAY;AAC7B,MAAmB,EACnB,YAAqB;IAErB,IAAI,YAAY,KAAK,IAAI,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY,IAAI,CAAC,EAAE,CAAC;QAC7E,OAAO,IAAI,CAAC,GAAG,CAAC,YAAY,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;IACnD,CAAC;IACD,MAAM,WAAW,GAAG,MAAM,CAAC,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;IAC9D,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;IACxD,kDAAkD;IAClD,OAAO,MAAM,GAAG,CAAC,GAAG,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,GAAG,CAAC,CAAC;AAC9C,CAAC;AAED,0DAA0D;AAC1D,MAAM,UAAU,iBAAiB,CAAC,MAAc;IAC9C,OAAO,MAAM,KAAK,GAAG,IAAI,MAAM,IAAI,GAAG,CAAC;AACzC,CAAC;AAED,uEAAuE;AACvE,MAAM,UAAU,YAAY,CAAC,IAAc;IACzC,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;IAC/C,IAAI,CAAC,MAAM;QAAE,OAAO,SAAS,CAAC;IAC9B,yCAAyC;IACzC,MAAM,IAAI,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC;IAChC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC;IAChD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IAChC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;IACxD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,+CAA+C;AAC/C,MAAM,UAAU,KAAK,CAAC,EAAU,EAAE,MAAoB;IACpD,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACpB,MAAM,CAAC,IAAI,YAAY,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC,CAAC;YAClD,OAAO;QACT,CAAC;QACD,MAAM,KAAK,GAAG,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACtC,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,GAAG,EAAE;YACrC,YAAY,CAAC,KAAK,CAAC,CAAC;YACpB,MAAM,CAAC,IAAI,YAAY,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC,CAAC;QACpD,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC"}

package/dist/telemetry.d.ts

@@ -0,0 +1,70 @@
+/**
+ * Telemetry hook types for the TensorCost Node SDK.
+ *
+ * Customers pass `onLifecycleEvent` in WrapOptions to receive typed events
+ * at each stage of a request lifecycle. The default is no-op — not providing
+ * the callback adds zero overhead to the hot path.
+ *
+ * What's included in every event:
+ *   - provider, model, operation — request identity
+ *   - attemptNumber             — 1-based; increments on each retry
+ *   - elapsedMs                 — wall-clock time since the first attempt began
+ *
+ * What is never included:
+ *   - prompt / completion content
+ *   - request or response bodies
+ *   - credentials
+ *
+ * The five event kinds map to the lifecycle stages:
+ *
+ *   before_request   fires before each attempt (including retried ones)
+ *   after_response   fires when a request completes successfully
+ *   on_retry         fires when a request is going to be retried, with the
+ *                    reason and how long the SDK will wait before the next
+ *                    attempt
+ *   on_error         fires when the SDK gives up (all retries exhausted or
+ *                    a non-retriable error occurred)
+ *   on_fallback      fires when the circuit breaker opens and the SDK routes
+ *                    to the provider directly instead of the proxy
+ */
+export type Provider = "openai" | "anthropic";
+/** Common fields shared by every lifecycle event. */
+export interface TelemetryBase {
+    provider: Provider;
+    model: string;
+    operation: string;
+    attemptNumber: number;
+    /** Wall-clock milliseconds since the first attempt for this call began. */
+    elapsedMs: number;
+}
+export interface BeforeRequestEvent extends TelemetryBase {
+    kind: "before_request";
+}
+export interface AfterResponseEvent extends TelemetryBase {
+    kind: "after_response";
+    status: number;
+    /** Request ID from the X-TC-Request-ID header, when the proxy set it. */
+    requestId: string | null;
+}
+export interface OnRetryEvent extends TelemetryBase {
+    kind: "on_retry";
+    reason: "5xx" | "network_error" | "quota_429";
+    status: number | null;
+    /** How many milliseconds the SDK will wait before the next attempt. */
+    delayMs: number;
+}
+export interface OnErrorEvent extends TelemetryBase {
+    kind: "on_error";
+    errorClass: string;
+    status: number | null;
+}
+export interface OnFallbackEvent extends TelemetryBase {
+    kind: "on_fallback";
+    /** Number of consecutive proxy failures that tripped the circuit. */
+    consecutiveFailures: number;
+}
+export type LifecycleEvent = BeforeRequestEvent | AfterResponseEvent | OnRetryEvent | OnErrorEvent | OnFallbackEvent;
+export type LifecycleEventCallback = (event: LifecycleEvent) => void;
+/** Call `cb` if defined — errors in the callback are always swallowed. */
+export declare function emitEvent(cb: LifecycleEventCallback | undefined, event: LifecycleEvent): void;
+//# sourceMappingURL=telemetry.d.ts.map

package/dist/telemetry.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"telemetry.d.ts","sourceRoot":"","sources":["../src/telemetry.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AAEH,MAAM,MAAM,QAAQ,GAAG,QAAQ,GAAG,WAAW,CAAC;AAE9C,qDAAqD;AACrD,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,QAAQ,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,aAAa,EAAE,MAAM,CAAC;IACtB,2EAA2E;IAC3E,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,kBAAmB,SAAQ,aAAa;IACvD,IAAI,EAAE,gBAAgB,CAAC;CACxB;AAED,MAAM,WAAW,kBAAmB,SAAQ,aAAa;IACvD,IAAI,EAAE,gBAAgB,CAAC;IACvB,MAAM,EAAE,MAAM,CAAC;IACf,yEAAyE;IACzE,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;CAC1B;AAED,MAAM,WAAW,YAAa,SAAQ,aAAa;IACjD,IAAI,EAAE,UAAU,CAAC;IACjB,MAAM,EAAE,KAAK,GAAG,eAAe,GAAG,WAAW,CAAC;IAC9C,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,uEAAuE;IACvE,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,YAAa,SAAQ,aAAa;IACjD,IAAI,EAAE,UAAU,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;CACvB;AAED,MAAM,WAAW,eAAgB,SAAQ,aAAa;IACpD,IAAI,EAAE,aAAa,CAAC;IACpB,qEAAqE;IACrE,mBAAmB,EAAE,MAAM,CAAC;CAC7B;AAED,MAAM,MAAM,cAAc,GACtB,kBAAkB,GAClB,kBAAkB,GAClB,YAAY,GACZ,YAAY,GACZ,eAAe,CAAC;AAEpB,MAAM,MAAM,sBAAsB,GAAG,CAAC,KAAK,EAAE,cAAc,KAAK,IAAI,CAAC;AAErE,0EAA0E;AAC1E,wBAAgB,SAAS,CACvB,EAAE,EAAE,sBAAsB,GAAG,SAAS,EACtC,KAAK,EAAE,cAAc,GACpB,IAAI,CAON"}

package/dist/telemetry.js

@@ -0,0 +1,41 @@
+/**
+ * Telemetry hook types for the TensorCost Node SDK.
+ *
+ * Customers pass `onLifecycleEvent` in WrapOptions to receive typed events
+ * at each stage of a request lifecycle. The default is no-op — not providing
+ * the callback adds zero overhead to the hot path.
+ *
+ * What's included in every event:
+ *   - provider, model, operation — request identity
+ *   - attemptNumber             — 1-based; increments on each retry
+ *   - elapsedMs                 — wall-clock time since the first attempt began
+ *
+ * What is never included:
+ *   - prompt / completion content
+ *   - request or response bodies
+ *   - credentials
+ *
+ * The five event kinds map to the lifecycle stages:
+ *
+ *   before_request   fires before each attempt (including retried ones)
+ *   after_response   fires when a request completes successfully
+ *   on_retry         fires when a request is going to be retried, with the
+ *                    reason and how long the SDK will wait before the next
+ *                    attempt
+ *   on_error         fires when the SDK gives up (all retries exhausted or
+ *                    a non-retriable error occurred)
+ *   on_fallback      fires when the circuit breaker opens and the SDK routes
+ *                    to the provider directly instead of the proxy
+ */
+/** Call `cb` if defined — errors in the callback are always swallowed. */
+export function emitEvent(cb, event) {
+    if (!cb)
+        return;
+    try {
+        cb(event);
+    }
+    catch {
+        // Customer callbacks must never blow up the SDK's control flow.
+    }
+}
+//# sourceMappingURL=telemetry.js.map

package/dist/telemetry.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"telemetry.js","sourceRoot":"","sources":["../src/telemetry.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AAsDH,0EAA0E;AAC1E,MAAM,UAAU,SAAS,CACvB,EAAsC,EACtC,KAAqB;IAErB,IAAI,CAAC,EAAE;QAAE,OAAO;IAChB,IAAI,CAAC;QACH,EAAE,CAAC,KAAK,CAAC,CAAC;IACZ,CAAC;IAAC,MAAM,CAAC;QACP,gEAAgE;IAClE,CAAC;AACH,CAAC"}

package/dist/types.d.ts

@@ -5,8 +5,10 @@
  * the customer's client type, so most provider shapes are duck-typed
  * at runtime rather than expressed in TypeScript.
  */
-export type Provider = "openai" | "anthropic";
-export type Operation = "chat.completions" | "completions" | "messages";
+import type { LifecycleEventCallback } from "./telemetry.js";
+import type { RetryConfig } from "./retry.js";
+export type Provider = "openai" | "anthropic" | "bedrock";
+export type Operation = "chat.completions" | "completions" | "messages" | "bedrock.invoke_model" | "bedrock.converse" | "bedrock.invoke_model_stream" | "bedrock.converse_stream";
 export type ObservationStatus = "success" | "error";
 /**
  * Wire shape posted to `POST /api/proxy/observation`. Mirrors the
@@ -112,6 +114,59 @@ export interface WrapOptions {
      * Example: `https://proxy.tensorcost.com`
      */
     proxyUrl?: string;
+    /**
+     * Retry configuration. Applies to applied-mode proxy requests only.
+     * Observe-only telemetry posts do not retry (they are fire-and-forget).
+     *
+     * Defaults: maxAttempts=5, baseDelayMs=500, maxDelayMs=30 000.
+     * Set `maxAttempts: 1` to disable retries entirely.
+     */
+    retry?: Partial<RetryConfig>;
+    /**
+     * Timeout for non-streaming proxy requests, in milliseconds.
+     * Default: 60 000 (60s). Surfaces as `TensorCostTimeoutError` (kind: "total").
+     */
+    timeoutMs?: number;
+    /**
+     * Idle timeout for streaming proxy requests (no bytes received for N ms).
+     * Default: 30 000 (30s). Surfaces as `TensorCostTimeoutError` (kind: "idle").
+     */
+    idleTimeoutMs?: number;
+    /**
+     * Optional lifecycle event callback. Fires on `before_request`,
+     * `after_response`, `on_retry`, `on_error`, and `on_fallback`.
+     * The payload contains only request metadata — never prompt / completion
+     * content or credentials. Errors in the callback are swallowed.
+     */
+    onLifecycleEvent?: LifecycleEventCallback;
+    /**
+     * Fail-open circuit breaker for the proxy. When the proxy returns 5xx
+     * for `openThreshold` consecutive calls (default 3), the SDK routes
+     * subsequent calls directly to the upstream provider, bypassing the
+     * proxy. The circuit closes again after `closeThreshold` consecutive
+     * successful probe calls (default 5).
+     *
+     * Default: true (enabled). Set to false when running in-VPC with a
+     * sidecar proxy where direct-provider access is not available.
+     *
+     * When false, proxy 5xx errors surface as `TensorCostProxyError` with
+     * no automatic fallback.
+     *
+     * The fallback requires that the customer's provider SDK client has its
+     * original credentials intact (pre-wrap). If the client credentials are
+     * not available, the SDK throws `TensorCostProxyError` instead.
+     */
+    failOpenEnabled?: boolean;
+    /**
+     * Provider API key used for the fail-open direct-provider fallback.
+     * Only relevant when `failOpenEnabled=true` and `appliedMode=true`.
+     *
+     * In most cases the SDK reads the provider key directly from the wrapped
+     * client (client.apiKey), so this is not needed. Supply it explicitly if
+     * the client exposes the key under a non-standard property name, or if
+     * you are using the SDK in a context where the client's key is unavailable.
+     */
+    providerApiKey?: string;
 }
 export interface ResolvedConfig {
     apiKey: string;
@@ -139,5 +194,11 @@ export interface ResolvedConfig {
      * a config with `appliedMode=true` and `proxyUrl=null`.
      */
     proxyUrl: string | null;
+    retry: RetryConfig;
+    timeoutMs: number;
+    idleTimeoutMs: number;
+    onLifecycleEvent: LifecycleEventCallback | undefined;
+    failOpenEnabled: boolean;
+    providerApiKey: string | null;
 }
 //# sourceMappingURL=types.d.ts.map

package/dist/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,MAAM,MAAM,QAAQ,GAAG,QAAQ,GAAG,WAAW,CAAC;AAE9C,MAAM,MAAM,SAAS,GAAG,kBAAkB,GAAG,aAAa,GAAG,UAAU,CAAC;AAExE,MAAM,MAAM,iBAAiB,GAAG,SAAS,GAAG,OAAO,CAAC;AAEpD;;;;GAIG;AACH,MAAM,WAAW,WAAW;IAC1B,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,QAAQ,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,SAAS,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,MAAM,EAAE,iBAAiB,CAAC;IAC1B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B;;;;;;OAMG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;;;;;;OAQG;IACH,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,WAAW;IAC1B,yEAAyE;IACzE,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,2EAA2E;IAC3E,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,2EAA2E;IAC3E,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;;;;;;OAQG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;;;;;;;;;;;;;;;OAiBG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;OAIG;IACH,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB;;;;;;;;;;;;;;OAcG;IACH,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB;;;;;;OAMG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,cAAc;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB;;;;;OAKG;IACH,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B;;;;OAIG;IACH,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,QAAQ,EAAE,OAAO,CAAC;IAClB,kCAAkC;IAClC,WAAW,EAAE,OAAO,CAAC;IACrB;;;;OAIG;IACH,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;CACzB"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,gBAAgB,CAAC;AAC7D,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAE9C,MAAM,MAAM,QAAQ,GAAG,QAAQ,GAAG,WAAW,GAAG,SAAS,CAAC;AAE1D,MAAM,MAAM,SAAS,GACjB,kBAAkB,GAClB,aAAa,GACb,UAAU,GACV,sBAAsB,GACtB,kBAAkB,GAClB,6BAA6B,GAC7B,yBAAyB,CAAC;AAE9B,MAAM,MAAM,iBAAiB,GAAG,SAAS,GAAG,OAAO,CAAC;AAEpD;;;;GAIG;AACH,MAAM,WAAW,WAAW;IAC1B,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,QAAQ,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,SAAS,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,MAAM,EAAE,iBAAiB,CAAC;IAC1B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B;;;;;;OAMG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;;;;;;OAQG;IACH,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,WAAW;IAC1B,yEAAyE;IACzE,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,2EAA2E;IAC3E,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,2EAA2E;IAC3E,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;;;;;;OAQG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;;;;;;;;;;;;;;;OAiBG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;OAIG;IACH,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB;;;;;;;;;;;;;;OAcG;IACH,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB;;;;;;OAMG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC;IAKlB;;;;;;OAMG;IACH,KAAK,CAAC,EAAE,OAAO,CAAC,WAAW,CAAC,CAAC;IAE7B;;;OAGG;IACH,SAAS,CAAC,EAAE,MAAM,CAAC;IAEnB;;;OAGG;IACH,aAAa,CAAC,EAAE,MAAM,CAAC;IAEvB;;;;;OAKG;IACH,gBAAgB,CAAC,EAAE,sBAAsB,CAAC;IAE1C;;;;;;;;;;;;;;;;OAgBG;IACH,eAAe,CAAC,EAAE,OAAO,CAAC;IAE1B;;;;;;;;OAQG;IACH,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,WAAW,cAAc;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB;;;;;OAKG;IACH,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B;;;;OAIG;IACH,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,QAAQ,EAAE,OAAO,CAAC;IAClB,kCAAkC;IAClC,WAAW,EAAE,OAAO,CAAC;IACrB;;;;OAIG;IACH,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IAGxB,KAAK,EAAE,WAAW,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,EAAE,sBAAsB,GAAG,SAAS,CAAC;IACrD,eAAe,EAAE,OAAO,CAAC;IACzB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;CAC/B"}

package/dist/wrap.d.ts

@@ -10,6 +10,15 @@
  * two custom headers (`x-tc-provider-url`, `x-tc-provider-auth`) are
  * injected into `client.defaultHeaders`. The proxy handles per-request
  * routing decisions. Observe-only telemetry still fires post-call.
+ *
+ * Hardening (v0.4.0): applied-mode requests now go through the proxy-client
+ * module which adds retries, timeouts, typed errors, telemetry hooks, and
+ * the fail-open circuit breaker.
+ *
+ * Bedrock (v0.4.0): observe-only. Applied mode is not supported for Bedrock —
+ * wrap(bedrockClient, { appliedMode: true }) throws TensorCostConfigError.
+ * The Smithy middleware registered on client.middlewareStack intercepts
+ * every client.send() call transparently.
  */
 import type { WrapOptions } from "./types.js";
 export declare function wrap<T>(client: T, options?: WrapOptions): T;

package/dist/wrap.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"wrap.d.ts","sourceRoot":"","sources":["../src/wrap.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAQH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAE9C,wBAAgB,IAAI,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,GAAE,WAAgB,GAAG,CAAC,CAyC/D"}
+{"version":3,"file":"wrap.d.ts","sourceRoot":"","sources":["../src/wrap.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAWH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAE9C,wBAAgB,IAAI,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,GAAE,WAAgB,GAAG,CAAC,CAqF/D"}