@pentoshi/clai 4.11.7 → 4.11.8

package/dist/llm/routing/key-rotation.js

@@ -0,0 +1,207 @@
+import { getActiveProviderEndpoint, getProviderEndpoints, providerUsesEndpoints, setActiveProviderEndpoint, } from "../../store/config.js";
+import { getProviderKeys, markProviderKeySuccess } from "../../store/keys.js";
+import { ProviderError } from "../http.js";
+import { attemptsPerKey, buildKeyAttemptPlan, formatKeyEventStatus, isImmediateKeySwitchError, isKeyCircleStopError, isKeyRotatableError, isQuotaKeyError, } from "../key-rotation.js";
+import { isOperationPolicyError } from "../operation-ledger.js";
+import { maskSecretTail } from "../provider.js";
+import { tryCompleteOnce } from "./attempt-complete.js";
+import { tryStreamOnce } from "./attempt-stream.js";
+import { isRateLimited, isRetriableError, isServerUnavailable, MAX_RETRIES, MAX_RETRY_WAIT_MS, networkRetryWaitMs, retryWaitMs, sleep, } from "./error-classification.js";
+import { authForSlot } from "./provider-selection.js";
+function failureReason(error) {
+    if (isRateLimited(error))
+        return "rate limited";
+    if (isQuotaKeyError(error)) {
+        const status = error instanceof ProviderError && error.status
+            ? ` (${error.status})`
+            : "";
+        return `insufficient credits${status}`;
+    }
+    if (error instanceof ProviderError && error.status) {
+        if (error.status === 401 || error.status === 403) {
+            return `auth failed (${error.status})`;
+        }
+        return `server error (${error.status})`;
+    }
+    return "connection glitch";
+}
+export function makeKeyEmitter(onStatus, onKeyEvent) {
+    return (event) => {
+        onKeyEvent?.(event);
+        if (event.type === "using")
+            return;
+        const line = formatKeyEventStatus(event);
+        if (event.type === "retry" ||
+            event.type === "switch" ||
+            event.type === "exhausted") {
+            onStatus?.(line);
+        }
+    };
+}
+export async function runWithKeyRotation(opts) {
+    const { providerId, provider, request, model, emitKey } = opts;
+    const singleDispatch = opts.singleDispatch === true;
+    const multi = await getProviderKeys(providerId);
+    const slots = multi.keys;
+    if (slots.length === 0) {
+        throw new Error("no API key configured");
+    }
+    const fullPlan = buildKeyAttemptPlan(slots.length, multi.activeIndex).filter((index) => !slots[index].disabled);
+    // large prompt again under a different credential.
+    const plan = singleDispatch ? fullPlan.slice(0, 1) : fullPlan;
+    if (plan.length === 0) {
+        throw new Error(`all ${slots.length} API key${slots.length === 1 ? "" : "s"} for ${providerId} are disabled — re-enable one to use this provider`);
+    }
+    const enabledCount = plan.length;
+    const multiKey = enabledCount > 1;
+    const maxPerKey = singleDispatch
+        ? 1
+        : attemptsPerKey(enabledCount, (opts.maxRetries ?? MAX_RETRIES) + 1);
+    let lastError;
+    const storedEndpoints = providerUsesEndpoints(providerId) && !singleDispatch
+        ? getProviderEndpoints(providerId)
+        : undefined;
+    const endpointUrls = (storedEndpoints?.urls ?? []).filter((url) => !(storedEndpoints?.disabledUrls ?? []).includes(url));
+    const activeEndpointUrl = storedEndpoints?.urls[storedEndpoints.activeIndex];
+    const activeEndpointPos = activeEndpointUrl
+        ? endpointUrls.indexOf(activeEndpointUrl)
+        : -1;
+    const endpointStart = activeEndpointPos >= 0 ? activeEndpointPos : 0;
+    let endpointOffset = 0;
+    const endpointCount = endpointUrls.length;
+    const authForAttempt = (value) => {
+        if (endpointCount === 0)
+            return authForSlot(providerId, value);
+        const url = endpointUrls[(endpointStart + endpointOffset) % endpointCount];
+        return { apiKey: value, baseUrl: url };
+    };
+    for (let planIdx = 0; planIdx < plan.length; planIdx++) {
+        const keyIndex = plan[planIdx];
+        const slot = slots[keyIndex];
+        const auth = authForAttempt(slot.value);
+        const tail = maskSecretTail(slot.value);
+        if (planIdx > 0) {
+            emitKey({
+                type: "switch",
+                provider: providerId,
+                maskedTail: tail,
+                keyIndex,
+                keyCount: slots.length,
+                ...(lastError ? { reason: failureReason(lastError) } : {}),
+            });
+        }
+        for (let attempt = 0; attempt < maxPerKey; attempt++) {
+            request.signal?.throwIfAborted();
+            const attemptReason = planIdx === 0 && attempt === 0 ? opts.initialAttemptReason : "retry";
+            try {
+                let result;
+                if (opts.mode === "stream") {
+                    result = await tryStreamOnce(provider, providerId, request, model, auth, opts.onToken ?? (() => { }), opts.onStatus, attemptReason, singleDispatch, opts.onSuccessfulRequest);
+                }
+                else {
+                    result = await tryCompleteOnce(provider, providerId, request, model, auth, attemptReason, opts.onStatus, singleDispatch);
+                }
+                if (multi.source !== "env" && multi.source !== "local") {
+                    void markProviderKeySuccess(providerId, keyIndex).catch(() => { });
+                }
+                if (storedEndpoints && endpointCount > 0) {
+                    const winningUrl = endpointUrls[(endpointStart + endpointOffset) % endpointCount];
+                    const storedIndex = storedEndpoints.urls.indexOf(winningUrl);
+                    const authoritative = getActiveProviderEndpoint(providerId);
+                    if (storedIndex >= 0 &&
+                        storedIndex !== storedEndpoints.activeIndex &&
+                        (authoritative === "" ||
+                            storedEndpoints.urls.includes(authoritative))) {
+                        try {
+                            setActiveProviderEndpoint(providerId, storedIndex);
+                        }
+                        catch { }
+                    }
+                }
+                return result;
+            }
+            catch (error) {
+                const previousError = lastError;
+                lastError = error;
+                if (isOperationPolicyError(error) && previousError !== undefined) {
+                    throw previousError;
+                }
+                if (isKeyCircleStopError(error)) {
+                    throw error;
+                }
+                if (isImmediateKeySwitchError(error)) {
+                    if (endpointCount > 1 && endpointOffset + 1 < endpointCount) {
+                        endpointOffset += 1;
+                        emitKey({
+                            type: "endpoint",
+                            provider: providerId,
+                            maskedTail: tail,
+                            reason: failureReason(error),
+                        });
+                        planIdx = -1;
+                        break;
+                    }
+                    if (multiKey) {
+                        break;
+                    }
+                    throw error;
+                }
+                const rotatable = isKeyRotatableError(error, isRetriableError);
+                if (!rotatable) {
+                    throw error;
+                }
+                const canRetrySame = attempt + 1 < maxPerKey;
+                if (canRetrySame) {
+                    const wait = isRateLimited(error) || isServerUnavailable(error)
+                        ? retryWaitMs(error, attempt)
+                        : networkRetryWaitMs(attempt);
+                    if (wait > MAX_RETRY_WAIT_MS) {
+                        if (multiKey)
+                            break;
+                        throw error;
+                    }
+                    emitKey({
+                        type: "retry",
+                        provider: providerId,
+                        maskedTail: tail,
+                        reason: failureReason(error),
+                        waitMs: wait,
+                        keyIndex,
+                        keyCount: slots.length,
+                    });
+                    await sleep(wait, request.signal);
+                    continue;
+                }
+                break;
+            }
+        }
+        if (planIdx >= 0 &&
+            endpointCount > 1 &&
+            planIdx + 1 < plan.length &&
+            lastError &&
+            !isImmediateKeySwitchError(lastError)) {
+            endpointOffset += 1;
+            emitKey({
+                type: "endpoint",
+                provider: providerId,
+                maskedTail: tail,
+                reason: failureReason(lastError),
+            });
+        }
+    }
+    if (!multiKey && lastError && isRateLimited(lastError)) {
+        opts.onStatus?.(`⏳ ${providerId} rate limited; staying on selected provider.`);
+    }
+    else if (multiKey) {
+        emitKey({
+            type: "exhausted",
+            provider: providerId,
+            maskedTail: maskSecretTail(slots[plan[0]].value),
+            keyCount: slots.length,
+            reason: lastError ? failureReason(lastError) : undefined,
+        });
+    }
+    const err = lastError ?? new Error("all API keys failed");
+    throw err;
+}
+//# sourceMappingURL=key-rotation.js.map

package/dist/llm/routing/key-rotation.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"key-rotation.js","sourceRoot":"","sources":["../../../src/llm/routing/key-rotation.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,yBAAyB,EACzB,oBAAoB,EACpB,qBAAqB,EACrB,yBAAyB,GAC1B,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAE,eAAe,EAAE,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AAO9E,OAAO,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAC3C,OAAO,EACL,cAAc,EACd,mBAAmB,EACnB,oBAAoB,EACpB,yBAAyB,EACzB,oBAAoB,EACpB,mBAAmB,EACnB,eAAe,GAChB,MAAM,oBAAoB,CAAC;AAE5B,OAAO,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AAChE,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAGhD,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EACL,aAAa,EACb,gBAAgB,EAChB,mBAAmB,EACnB,WAAW,EACX,iBAAiB,EACjB,kBAAkB,EAClB,WAAW,EACX,KAAK,GACN,MAAM,2BAA2B,CAAC;AACnC,OAAO,EAAE,WAAW,EAAE,MAAM,yBAAyB,CAAC;AAEtD,SAAS,aAAa,CAAC,KAAc;IACnC,IAAI,aAAa,CAAC,KAAK,CAAC;QAAE,OAAO,cAAc,CAAC;IAChD,IAAI,eAAe,CAAC,KAAK,CAAC,EAAE,CAAC;QAC3B,MAAM,MAAM,GACV,KAAK,YAAY,aAAa,IAAI,KAAK,CAAC,MAAM;YAC5C,CAAC,CAAC,KAAK,KAAK,CAAC,MAAM,GAAG;YACtB,CAAC,CAAC,EAAE,CAAC;QACT,OAAO,uBAAuB,MAAM,EAAE,CAAC;IACzC,CAAC;IACD,IAAI,KAAK,YAAY,aAAa,IAAI,KAAK,CAAC,MAAM,EAAE,CAAC;QACnD,IAAI,KAAK,CAAC,MAAM,KAAK,GAAG,IAAI,KAAK,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;YACjD,OAAO,gBAAgB,KAAK,CAAC,MAAM,GAAG,CAAC;QACzC,CAAC;QACD,OAAO,iBAAiB,KAAK,CAAC,MAAM,GAAG,CAAC;IAC1C,CAAC;IACD,OAAO,mBAAmB,CAAC;AAC7B,CAAC;AAID,MAAM,UAAU,cAAc,CAC5B,QAAoC,EACpC,UAAoC;IAEpC,OAAO,CAAC,KAAK,EAAE,EAAE;QACf,UAAU,EAAE,CAAC,KAAK,CAAC,CAAC;QACpB,IAAI,KAAK,CAAC,IAAI,KAAK,OAAO;YAAE,OAAO;QACnC,MAAM,IAAI,GAAG,oBAAoB,CAAC,KAAK,CAAC,CAAC;QACzC,IACE,KAAK,CAAC,IAAI,KAAK,OAAO;YACtB,KAAK,CAAC,IAAI,KAAK,QAAQ;YACvB,KAAK,CAAC,IAAI,KAAK,WAAW,EAC1B,CAAC;YACD,QAAQ,EAAE,CAAC,IAAI,CAAC,CAAC;QACnB,CAAC;IACH,CAAC,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAI,IAc3C;IACC,MAAM,EAAE,UAAU,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC;IAC/D,MAAM,cAAc,GAAG,IAAI,CAAC,cAAc,KAAK,IAAI,CAAC;IACpD,MAAM,KAAK,GAAG,MAAM,eAAe,CAAC,UAAU,CAAC,CAAC;IAChD,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC;IACzB,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,MAAM,IAAI,KAAK,CAAC,uBAAuB,CAAC,CAAC;IAC3C,CAAC;IAED,MAAM,QAAQ,GAAG,mBAAmB,CAAC,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,WAAW,CAAC,CAAC,MAAM,CAC1E,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,KAAK,CAAE,CAAC,QAAQ,CACnC,CAAC;IACF,mDAAmD;IACnD,MAAM,IAAI,GAAG,cAAc,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC;IAC9D,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CACb,OAAO,KAAK,CAAC,MAAM,WAAW,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,QAAQ,UAAU,oDAAoD,CAClI,CAAC;IACJ,CAAC;IACD,MAAM,YAAY,GAAG,IAAI,CAAC,MAAM,CAAC;IACjC,MAAM,QAAQ,GAAG,YAAY,GAAG,CAAC,CAAC;IAClC,MAAM,SAAS,GAAG,cAAc;QAC9B,CAAC,CAAC,CAAC;QACH,CAAC,CAAC,cAAc,CAAC,YAAY,EAAE,CAAC,IAAI,CAAC,UAAU,IAAI,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;IACvE,IAAI,SAAkB,CAAC;IAEvB,MAAM,eAAe,GACnB,qBAAqB,CAAC,UAAU,CAAC,IAAI,CAAC,cAAc;QAClD,CAAC,CAAC,oBAAoB,CAAC,UAAU,CAAC;QAClC,CAAC,CAAC,SAAS,CAAC;IAChB,MAAM,YAAY,GAAG,CAAC,eAAe,EAAE,IAAI,IAAI,EAAE,CAAC,CAAC,MAAM,CACvD,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,CAAC,eAAe,EAAE,YAAY,IAAI,EAAE,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,CAC9D,CAAC;IACF,MAAM,iBAAiB,GAAG,eAAe,EAAE,IAAI,CAAC,eAAe,CAAC,WAAW,CAAC,CAAC;IAC7E,MAAM,iBAAiB,GAAG,iBAAiB;QACzC,CAAC,CAAC,YAAY,CAAC,OAAO,CAAC,iBAAiB,CAAC;QACzC,CAAC,CAAC,CAAC,CAAC,CAAC;IACP,MAAM,aAAa,GAAG,iBAAiB,IAAI,CAAC,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,CAAC,CAAC;IACrE,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,MAAM,aAAa,GAAG,YAAY,CAAC,MAAM,CAAC;IAC1C,MAAM,cAAc,GAAG,CAAC,KAAyB,EAAgB,EAAE;QACjE,IAAI,aAAa,KAAK,CAAC;YAAE,OAAO,WAAW,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC;QAC/D,MAAM,GAAG,GAAG,YAAY,CAAC,CAAC,aAAa,GAAG,cAAc,CAAC,GAAG,aAAa,CAAE,CAAC;QAC5E,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC;IACzC,CAAC,CAAC;IAEF,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,IAAI,CAAC,MAAM,EAAE,OAAO,EAAE,EAAE,CAAC;QACvD,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,CAAE,CAAC;QAChC,MAAM,IAAI,GAAG,KAAK,CAAC,QAAQ,CAAE,CAAC;QAC9B,MAAM,IAAI,GAAG,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACxC,MAAM,IAAI,GAAG,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAExC,IAAI,OAAO,GAAG,CAAC,EAAE,CAAC;YAChB,OAAO,CAAC;gBACN,IAAI,EAAE,QAAQ;gBACd,QAAQ,EAAE,UAAU;gBACpB,UAAU,EAAE,IAAI;gBAChB,QAAQ;gBACR,QAAQ,EAAE,KAAK,CAAC,MAAM;gBACtB,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,aAAa,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC3D,CAAC,CAAC;QACL,CAAC;QAED,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,SAAS,EAAE,OAAO,EAAE,EAAE,CAAC;YACrD,OAAO,CAAC,MAAM,EAAE,cAAc,EAAE,CAAC;YACjC,MAAM,aAAa,GACjB,OAAO,KAAK,CAAC,IAAI,OAAO,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC,CAAC,OAAO,CAAC;YACvE,IAAI,CAAC;gBACH,IAAI,MAAwB,CAAC;gBAC7B,IAAI,IAAI,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;oBAC3B,MAAM,GAAG,MAAM,aAAa,CAC1B,QAAQ,EACR,UAAU,EACV,OAAO,EACP,KAAK,EACL,IAAI,EACJ,IAAI,CAAC,OAAO,IAAI,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,EAC1B,IAAI,CAAC,QAAQ,EACb,aAAa,EACb,cAAc,EACd,IAAI,CAAC,mBAAmB,CACzB,CAAC;gBACJ,CAAC;qBAAM,CAAC;oBACN,MAAM,GAAG,MAAM,eAAe,CAC5B,QAAQ,EACR,UAAU,EACV,OAAO,EACP,KAAK,EACL,IAAI,EACJ,aAAa,EACb,IAAI,CAAC,QAAQ,EACb,cAAc,CACf,CAAC;gBACJ,CAAC;gBACD,IAAI,KAAK,CAAC,MAAM,KAAK,KAAK,IAAI,KAAK,CAAC,MAAM,KAAK,OAAO,EAAE,CAAC;oBACvD,KAAK,sBAAsB,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;gBACpE,CAAC;gBACD,IAAI,eAAe,IAAI,aAAa,GAAG,CAAC,EAAE,CAAC;oBACzC,MAAM,UAAU,GACd,YAAY,CAAC,CAAC,aAAa,GAAG,cAAc,CAAC,GAAG,aAAa,CAAE,CAAC;oBAClE,MAAM,WAAW,GAAG,eAAe,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;oBAC7D,MAAM,aAAa,GAAG,yBAAyB,CAAC,UAAU,CAAC,CAAC;oBAC5D,IACE,WAAW,IAAI,CAAC;wBAChB,WAAW,KAAK,eAAe,CAAC,WAAW;wBAC3C,CAAC,aAAa,KAAK,EAAE;4BACnB,eAAe,CAAC,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,CAAC,EAC/C,CAAC;wBACD,IAAI,CAAC;4BACH,yBAAyB,CAAC,UAAU,EAAE,WAAW,CAAC,CAAC;wBACrD,CAAC;wBAAC,MAAM,CAAC,CAAA,CAAC;oBACZ,CAAC;gBACH,CAAC;gBACD,OAAO,MAAW,CAAC;YACrB,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,MAAM,aAAa,GAAG,SAAS,CAAC;gBAChC,SAAS,GAAG,KAAK,CAAC;gBAElB,IAAI,sBAAsB,CAAC,KAAK,CAAC,IAAI,aAAa,KAAK,SAAS,EAAE,CAAC;oBACjE,MAAM,aAAa,CAAC;gBACtB,CAAC;gBAED,IAAI,oBAAoB,CAAC,KAAK,CAAC,EAAE,CAAC;oBAChC,MAAM,KAAK,CAAC;gBACd,CAAC;gBAED,IAAI,yBAAyB,CAAC,KAAK,CAAC,EAAE,CAAC;oBACrC,IAAI,aAAa,GAAG,CAAC,IAAI,cAAc,GAAG,CAAC,GAAG,aAAa,EAAE,CAAC;wBAC5D,cAAc,IAAI,CAAC,CAAC;wBACpB,OAAO,CAAC;4BACN,IAAI,EAAE,UAAU;4BAChB,QAAQ,EAAE,UAAU;4BACpB,UAAU,EAAE,IAAI;4BAChB,MAAM,EAAE,aAAa,CAAC,KAAK,CAAC;yBAC7B,CAAC,CAAC;wBACH,OAAO,GAAG,CAAC,CAAC,CAAC;wBACb,MAAM;oBACR,CAAC;oBACD,IAAI,QAAQ,EAAE,CAAC;wBACb,MAAM;oBACR,CAAC;oBACD,MAAM,KAAK,CAAC;gBACd,CAAC;gBAED,MAAM,SAAS,GAAG,mBAAmB,CAAC,KAAK,EAAE,gBAAgB,CAAC,CAAC;gBAC/D,IAAI,CAAC,SAAS,EAAE,CAAC;oBACf,MAAM,KAAK,CAAC;gBACd,CAAC;gBAED,MAAM,YAAY,GAAG,OAAO,GAAG,CAAC,GAAG,SAAS,CAAC;gBAC7C,IAAI,YAAY,EAAE,CAAC;oBACjB,MAAM,IAAI,GACR,aAAa,CAAC,KAAK,CAAC,IAAI,mBAAmB,CAAC,KAAK,CAAC;wBAChD,CAAC,CAAC,WAAW,CAAC,KAAK,EAAE,OAAO,CAAC;wBAC7B,CAAC,CAAC,kBAAkB,CAAC,OAAO,CAAC,CAAC;oBAClC,IAAI,IAAI,GAAG,iBAAiB,EAAE,CAAC;wBAC7B,IAAI,QAAQ;4BAAE,MAAM;wBACpB,MAAM,KAAK,CAAC;oBACd,CAAC;oBACD,OAAO,CAAC;wBACN,IAAI,EAAE,OAAO;wBACb,QAAQ,EAAE,UAAU;wBACpB,UAAU,EAAE,IAAI;wBAChB,MAAM,EAAE,aAAa,CAAC,KAAK,CAAC;wBAC5B,MAAM,EAAE,IAAI;wBACZ,QAAQ;wBACR,QAAQ,EAAE,KAAK,CAAC,MAAM;qBACvB,CAAC,CAAC;oBACH,MAAM,KAAK,CAAC,IAAI,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;oBAClC,SAAS;gBACX,CAAC;gBACD,MAAM;YACR,CAAC;QACH,CAAC;QAED,IACE,OAAO,IAAI,CAAC;YACZ,aAAa,GAAG,CAAC;YACjB,OAAO,GAAG,CAAC,GAAG,IAAI,CAAC,MAAM;YACzB,SAAS;YACT,CAAC,yBAAyB,CAAC,SAAS,CAAC,EACrC,CAAC;YACD,cAAc,IAAI,CAAC,CAAC;YACpB,OAAO,CAAC;gBACN,IAAI,EAAE,UAAU;gBAChB,QAAQ,EAAE,UAAU;gBACpB,UAAU,EAAE,IAAI;gBAChB,MAAM,EAAE,aAAa,CAAC,SAAS,CAAC;aACjC,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,IAAI,CAAC,QAAQ,IAAI,SAAS,IAAI,aAAa,CAAC,SAAS,CAAC,EAAE,CAAC;QACvD,IAAI,CAAC,QAAQ,EAAE,CACb,KAAK,UAAU,8CAA8C,CAC9D,CAAC;IACJ,CAAC;SAAM,IAAI,QAAQ,EAAE,CAAC;QACpB,OAAO,CAAC;YACN,IAAI,EAAE,WAAW;YACjB,QAAQ,EAAE,UAAU;YACpB,UAAU,EAAE,cAAc,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAE,CAAE,CAAC,KAAK,CAAC;YAClD,QAAQ,EAAE,KAAK,CAAC,MAAM;YACtB,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC,aAAa,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS;SACzD,CAAC,CAAC;IACL,CAAC;IAED,MAAM,GAAG,GAAG,SAAS,IAAI,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;IAC1D,MAAM,GAAG,CAAC;AACZ,CAAC"}

package/dist/llm/routing/provider-selection.d.ts

@@ -0,0 +1,8 @@
+import type { ProviderId } from "../../types.js";
+import type { LlmProvider, ProviderAuth } from "../provider.js";
+export declare const providers: Record<ProviderId, LlmProvider>;
+export declare function requestedRealKeyCount(provider: ProviderId): Promise<number>;
+export declare function buildFallbackChain(requested: ProviderId, freeOnly: boolean, enabled?: boolean, preferAlternates?: boolean): ProviderId[];
+export declare function getProvider(provider: ProviderId): LlmProvider;
+export declare function providerAuth(provider: ProviderId): Promise<ProviderAuth>;
+export declare function authForSlot(providerId: ProviderId, value: string | undefined): ProviderAuth;

package/dist/llm/routing/provider-selection.js

@@ -0,0 +1,119 @@
+import { getActiveProviderEndpoint, getCustomProviders, providerUsesEndpoints, resolveProviderCategory, } from "../../store/config.js";
+import { getProviderKeys, getProviderSecret } from "../../store/keys.js";
+import { agentrouterProvider } from "../agentrouter.js";
+import { anthropicProvider } from "../anthropic.js";
+import { mantleProvider } from "../aws-mantle.js";
+import { bynaraProvider } from "../bynara.js";
+import { getCustomProviderSync } from "../custom-providers.js";
+import { fireworksProvider } from "../fireworks.js";
+import { freeProvider } from "../free.js";
+import { geminiProvider } from "../gemini.js";
+import { hetznerProvider } from "../hetzner.js";
+import { lightningProvider } from "../lightning.js";
+import { mergeGatewayProvider } from "../merge-gateway.js";
+import { metaProvider } from "../meta.js";
+import { modalProvider } from "../modal.js";
+import { nvidiaProvider } from "../nvidia.js";
+import { ollamaProvider } from "../ollama.js";
+import { openaiProvider } from "../openai.js";
+import { openrouterProvider } from "../openrouter.js";
+import { orcarouterProvider } from "../orcarouter.js";
+import { qwenCloudProvider } from "../qwen-cloud.js";
+import { tokenrouterProvider } from "../tokenrouter.js";
+export const providers = {
+    free: freeProvider,
+    gemini: geminiProvider,
+    openrouter: openrouterProvider,
+    openai: openaiProvider,
+    anthropic: anthropicProvider,
+    nvidia: nvidiaProvider,
+    agentrouter: agentrouterProvider,
+    "aws-mantle": mantleProvider,
+    ollama: ollamaProvider,
+    bynara: bynaraProvider,
+    "qwen-cloud": qwenCloudProvider,
+    modal: modalProvider,
+    lightning: lightningProvider,
+    tokenrouter: tokenrouterProvider,
+    meta: metaProvider,
+    fireworks: fireworksProvider,
+    hetzner: hetznerProvider,
+    orcarouter: orcarouterProvider,
+    "merge-gateway": mergeGatewayProvider,
+};
+const fallbackOrder = [
+    "free",
+    "nvidia",
+    "gemini",
+    "openrouter",
+    "agentrouter",
+    "bynara",
+    "openai",
+    "anthropic",
+    "aws-mantle",
+    "ollama",
+    "qwen-cloud",
+    "modal",
+    "lightning",
+    "tokenrouter",
+    "meta",
+    "fireworks",
+    "hetzner",
+    "orcarouter",
+    "merge-gateway",
+];
+function allFallbackIds() {
+    const custom = getCustomProviders().map((d) => d.id);
+    return [...fallbackOrder, ...custom];
+}
+export async function requestedRealKeyCount(provider) {
+    // Ollama's "key" is a local host URL, and `free` has no credential at all —
+    if (provider === "ollama" || provider === "free")
+        return 0;
+    const multi = await getProviderKeys(provider);
+    return multi.keys.filter((key) => key.value && !key.disabled).length;
+}
+export function buildFallbackChain(requested, freeOnly, enabled = false, preferAlternates = false) {
+    if (!enabled)
+        return [requested];
+    const order = allFallbackIds();
+    const filtered = freeOnly
+        ? order.filter((provider) => provider === requested ||
+            resolveProviderCategory(provider) !== "paid-cloud")
+        : order;
+    const alternates = filtered.filter((provider) => provider !== requested);
+    return preferAlternates
+        ? [...alternates, requested]
+        : [requested, ...alternates];
+}
+export function getProvider(provider) {
+    const builtin = providers[provider];
+    if (builtin)
+        return builtin;
+    const custom = getCustomProviderSync(provider);
+    if (custom)
+        return custom;
+    return providers.nvidia;
+}
+export async function providerAuth(provider) {
+    const secret = await getProviderSecret(provider);
+    if (provider === "ollama") {
+        return { baseUrl: secret.value };
+    }
+    if (providerUsesEndpoints(provider)) {
+        const baseUrl = getActiveProviderEndpoint(provider);
+        return { apiKey: secret.value, ...(baseUrl ? { baseUrl } : {}) };
+    }
+    return { apiKey: secret.value };
+}
+export function authForSlot(providerId, value) {
+    if (providerId === "ollama") {
+        return { baseUrl: value };
+    }
+    if (providerUsesEndpoints(providerId)) {
+        const baseUrl = getActiveProviderEndpoint(providerId);
+        return { apiKey: value, ...(baseUrl ? { baseUrl } : {}) };
+    }
+    return { apiKey: value };
+}
+//# sourceMappingURL=provider-selection.js.map

package/dist/llm/routing/provider-selection.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"provider-selection.js","sourceRoot":"","sources":["../../../src/llm/routing/provider-selection.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,yBAAyB,EACzB,kBAAkB,EAClB,qBAAqB,EACrB,uBAAuB,GACxB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAE,eAAe,EAAE,iBAAiB,EAAE,MAAM,qBAAqB,CAAC;AAEzE,OAAO,EAAE,mBAAmB,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,iBAAiB,EAAE,MAAM,iBAAiB,CAAC;AACpD,OAAO,EAAE,cAAc,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC9C,OAAO,EAAE,qBAAqB,EAAE,MAAM,wBAAwB,CAAC;AAC/D,OAAO,EAAE,iBAAiB,EAAE,MAAM,iBAAiB,CAAC;AACpD,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAC1C,OAAO,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC9C,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAChD,OAAO,EAAE,iBAAiB,EAAE,MAAM,iBAAiB,CAAC;AACpD,OAAO,EAAE,oBAAoB,EAAE,MAAM,qBAAqB,CAAC;AAC3D,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAC1C,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAC5C,OAAO,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC9C,OAAO,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC9C,OAAO,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC9C,OAAO,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AACtD,OAAO,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AAEtD,OAAO,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AACrD,OAAO,EAAE,mBAAmB,EAAE,MAAM,mBAAmB,CAAC;AAExD,MAAM,CAAC,MAAM,SAAS,GAAoC;IACxD,IAAI,EAAE,YAAY;IAClB,MAAM,EAAE,cAAc;IACtB,UAAU,EAAE,kBAAkB;IAC9B,MAAM,EAAE,cAAc;IACtB,SAAS,EAAE,iBAAiB;IAC5B,MAAM,EAAE,cAAc;IACtB,WAAW,EAAE,mBAAmB;IAChC,YAAY,EAAE,cAAc;IAC5B,MAAM,EAAE,cAAc;IACtB,MAAM,EAAE,cAAc;IACtB,YAAY,EAAE,iBAAiB;IAC/B,KAAK,EAAE,aAAa;IACpB,SAAS,EAAE,iBAAiB;IAC5B,WAAW,EAAE,mBAAmB;IAChC,IAAI,EAAE,YAAY;IAClB,SAAS,EAAE,iBAAiB;IAC5B,OAAO,EAAE,eAAe;IACxB,UAAU,EAAE,kBAAkB;IAC9B,eAAe,EAAE,oBAAoB;CACtC,CAAC;AAEF,MAAM,aAAa,GAAiB;IAClC,MAAM;IACN,QAAQ;IACR,QAAQ;IACR,YAAY;IACZ,aAAa;IACb,QAAQ;IACR,QAAQ;IACR,WAAW;IACX,YAAY;IACZ,QAAQ;IACR,YAAY;IACZ,OAAO;IACP,WAAW;IACX,aAAa;IACb,MAAM;IACN,WAAW;IACX,SAAS;IACT,YAAY;IACZ,eAAe;CAChB,CAAC;AAEF,SAAS,cAAc;IACrB,MAAM,MAAM,GAAG,kBAAkB,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAgB,CAAC,CAAC;IACnE,OAAO,CAAC,GAAG,aAAa,EAAE,GAAG,MAAM,CAAC,CAAC;AACvC,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,qBAAqB,CACzC,QAAoB;IAEpB,4EAA4E;IAC5E,IAAI,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,MAAM;QAAE,OAAO,CAAC,CAAC;IAC3D,MAAM,KAAK,GAAG,MAAM,eAAe,CAAC,QAAQ,CAAC,CAAC;IAC9C,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC;AACvE,CAAC;AAED,MAAM,UAAU,kBAAkB,CAChC,SAAqB,EACrB,QAAiB,EACjB,OAAO,GAAG,KAAK,EACf,gBAAgB,GAAG,KAAK;IAExB,IAAI,CAAC,OAAO;QAAE,OAAO,CAAC,SAAS,CAAC,CAAC;IACjC,MAAM,KAAK,GAAG,cAAc,EAAE,CAAC;IAC/B,MAAM,QAAQ,GAAG,QAAQ;QACvB,CAAC,CAAC,KAAK,CAAC,MAAM,CACV,CAAC,QAAQ,EAAE,EAAE,CACX,QAAQ,KAAK,SAAS;YACtB,uBAAuB,CAAC,QAAQ,CAAC,KAAK,YAAY,CACrD;QACH,CAAC,CAAC,KAAK,CAAC;IACV,MAAM,UAAU,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC;IACzE,OAAO,gBAAgB;QACrB,CAAC,CAAC,CAAC,GAAG,UAAU,EAAE,SAAS,CAAC;QAC5B,CAAC,CAAC,CAAC,SAAS,EAAE,GAAG,UAAU,CAAC,CAAC;AACjC,CAAC;AAED,MAAM,UAAU,WAAW,CAAC,QAAoB;IAC9C,MAAM,OAAO,GAAG,SAAS,CAAC,QAAQ,CAAC,CAAC;IACpC,IAAI,OAAO;QAAE,OAAO,OAAO,CAAC;IAC5B,MAAM,MAAM,GAAG,qBAAqB,CAAC,QAAkB,CAAC,CAAC;IACzD,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAC1B,OAAO,SAAS,CAAC,MAAM,CAAC;AAC1B,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,QAAoB;IAEpB,MAAM,MAAM,GAAG,MAAM,iBAAiB,CAAC,QAAQ,CAAC,CAAC;IACjD,IAAI,QAAQ,KAAK,QAAQ,EAAE,CAAC;QAC1B,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,KAAK,EAAE,CAAC;IACnC,CAAC;IACD,IAAI,qBAAqB,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,MAAM,OAAO,GAAG,yBAAyB,CAAC,QAAQ,CAAC,CAAC;QACpD,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC;IACnE,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,EAAE,CAAC;AAClC,CAAC;AAED,MAAM,UAAU,WAAW,CACzB,UAAsB,EACtB,KAAyB;IAEzB,IAAI,UAAU,KAAK,QAAQ,EAAE,CAAC;QAC5B,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;IAC5B,CAAC;IACD,IAAI,qBAAqB,CAAC,UAAU,CAAC,EAAE,CAAC;QACtC,MAAM,OAAO,GAAG,yBAAyB,CAAC,UAAU,CAAC,CAAC;QACtD,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC;IAC5D,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;AAC3B,CAAC"}

package/dist/llm/sampling.d.ts

@@ -1,15 +1,4 @@
 import type { ProviderId } from "../types.js";
-/**
- * One declarative sampling policy.
- *
- * Everything used to run at `temperature: 0.2` with a single hard-coded MiniMax
- * exception and no `top_p` plumbing at all. Reasoning families degrade badly
- * under near-greedy decoding (Qwen3 in thinking mode loops, DeepSeek-R1 wants
- * ~0.6, gpt-oss wants 1.0), which is exactly the set of models users enable
- * thinking for.
- *
- * Add model families here, not as another regex inside an adapter.
- */
 export interface SamplingDefaults {
     temperature: number;
     topP?: number | undefined;
@@ -20,10 +9,6 @@ export declare function samplingDefaults(input: {
     model: string;
     reasoningEnabled?: boolean | undefined;
 }): SamplingDefaults;
-/**
- * Effective sampling for a request: an explicit caller/user temperature always
- * wins; `top_p` is only sent when the policy asks for it.
- */
 export declare function resolveSampling(input: {
     provider?: ProviderId | undefined;
     model: string;

package/dist/llm/sampling.js

@@ -1,15 +1,10 @@
 export const DEFAULT_SAMPLING = { temperature: 0.2 };
-/** First match wins; order from most specific to least. */
 const SAMPLING_RULES = [
-    // Kimchi exposes this as `minimax-m3`; NVIDIA uses `minimaxai/minimax-m3`.
-    // Both require near-full sampling or the model degenerates.
     {
         pattern: /minimax-m3/i,
         defaults: { temperature: 1.0, topP: 0.95 },
     },
-    // OpenAI's gpt-oss weights are trained for temperature 1.0.
     { pattern: /gpt-oss/i, defaults: { temperature: 1.0 } },
-    // Thinking-mode recommendations from the model cards.
     {
         pattern: /qwen-?3|qwen3/i,
         reasoningOnly: true,
@@ -34,10 +29,6 @@ export function samplingDefaults(input) {
     }
     return DEFAULT_SAMPLING;
 }
-/**
- * Effective sampling for a request: an explicit caller/user temperature always
- * wins; `top_p` is only sent when the policy asks for it.
- */
 export function resolveSampling(input) {
     const policy = samplingDefaults(input);
     return {

package/dist/llm/sampling.js.map

@@ -1 +1 @@
-{"version":3,"file":"sampling.js","sourceRoot":"","sources":["../../src/llm/sampling.ts"],"names":[],"mappings":"AA4BA,MAAM,CAAC,MAAM,gBAAgB,GAAqB,EAAE,WAAW,EAAE,GAAG,EAAE,CAAC;AAEvE,2DAA2D;AAC3D,MAAM,cAAc,GAAmB;IACrC,2EAA2E;IAC3E,4DAA4D;IAC5D;QACE,OAAO,EAAE,aAAa;QACtB,QAAQ,EAAE,EAAE,WAAW,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE;KAC3C;IACD,4DAA4D;IAC5D,EAAE,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,EAAE,WAAW,EAAE,GAAG,EAAE,EAAE;IACvD,sDAAsD;IACtD;QACE,OAAO,EAAE,gBAAgB;QACzB,aAAa,EAAE,IAAI;QACnB,QAAQ,EAAE,EAAE,WAAW,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE;KAC3C;IACD;QACE,OAAO,EAAE,0BAA0B;QACnC,aAAa,EAAE,IAAI;QACnB,QAAQ,EAAE,EAAE,WAAW,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE;KAC3C;CACF,CAAC;AAEF,MAAM,UAAU,gBAAgB,CAAC,KAIhC;IACC,KAAK,MAAM,IAAI,IAAI,cAAc,EAAE,CAAC;QAClC,IAAI,IAAI,CAAC,aAAa,IAAI,CAAC,KAAK,CAAC,gBAAgB;YAAE,SAAS;QAC5D,IACE,IAAI,CAAC,SAAS;YACd,CAAC,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,EAC7D,CAAC;YACD,SAAS;QACX,CAAC;QACD,IAAI,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC,QAAQ,CAAC;IAC3D,CAAC;IACD,OAAO,gBAAgB,CAAC;AAC1B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,KAK/B;IACC,MAAM,MAAM,GAAG,gBAAgB,CAAC,KAAK,CAAC,CAAC;IACvC,OAAO;QACL,WAAW,EAAE,KAAK,CAAC,oBAAoB,IAAI,MAAM,CAAC,WAAW;QAC7D,GAAG,CAAC,MAAM,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC5D,CAAC;AACJ,CAAC"}
+{"version":3,"file":"sampling.js","sourceRoot":"","sources":["../../src/llm/sampling.ts"],"names":[],"mappings":"AAcA,MAAM,CAAC,MAAM,gBAAgB,GAAqB,EAAE,WAAW,EAAE,GAAG,EAAE,CAAC;AAEvE,MAAM,cAAc,GAAmB;IACrC;QACE,OAAO,EAAE,aAAa;QACtB,QAAQ,EAAE,EAAE,WAAW,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE;KAC3C;IACD,EAAE,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,EAAE,WAAW,EAAE,GAAG,EAAE,EAAE;IACvD;QACE,OAAO,EAAE,gBAAgB;QACzB,aAAa,EAAE,IAAI;QACnB,QAAQ,EAAE,EAAE,WAAW,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE;KAC3C;IACD;QACE,OAAO,EAAE,0BAA0B;QACnC,aAAa,EAAE,IAAI;QACnB,QAAQ,EAAE,EAAE,WAAW,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE;KAC3C;CACF,CAAC;AAEF,MAAM,UAAU,gBAAgB,CAAC,KAIhC;IACC,KAAK,MAAM,IAAI,IAAI,cAAc,EAAE,CAAC;QAClC,IAAI,IAAI,CAAC,aAAa,IAAI,CAAC,KAAK,CAAC,gBAAgB;YAAE,SAAS;QAC5D,IACE,IAAI,CAAC,SAAS;YACd,CAAC,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,EAC7D,CAAC;YACD,SAAS;QACX,CAAC;QACD,IAAI,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC,QAAQ,CAAC;IAC3D,CAAC;IACD,OAAO,gBAAgB,CAAC;AAC1B,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,KAK/B;IACC,MAAM,MAAM,GAAG,gBAAgB,CAAC,KAAK,CAAC,CAAC;IACvC,OAAO;QACL,WAAW,EAAE,KAAK,CAAC,oBAAoB,IAAI,MAAM,CAAC,WAAW;QAC7D,GAAG,CAAC,MAAM,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC5D,CAAC;AACJ,CAAC"}

package/dist/llm/session-affinity.d.ts

@@ -0,0 +1,2 @@
+export declare function withSessionAffinity<T>(sessionId: string, run: () => T): T;
+export declare function currentSessionAffinity(): string | undefined;

package/dist/llm/session-affinity.js

@@ -0,0 +1,9 @@
+import { AsyncLocalStorage } from "node:async_hooks";
+const sessionAffinityStorage = new AsyncLocalStorage();
+export function withSessionAffinity(sessionId, run) {
+    return sessionAffinityStorage.run(sessionId, run);
+}
+export function currentSessionAffinity() {
+    return sessionAffinityStorage.getStore();
+}
+//# sourceMappingURL=session-affinity.js.map

package/dist/llm/session-affinity.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"session-affinity.js","sourceRoot":"","sources":["../../src/llm/session-affinity.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAErD,MAAM,sBAAsB,GAAG,IAAI,iBAAiB,EAAU,CAAC;AAE/D,MAAM,UAAU,mBAAmB,CACjC,SAAiB,EACjB,GAAY;IAEZ,OAAO,sBAAsB,CAAC,GAAG,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC;AACpD,CAAC;AAED,MAAM,UAAU,sBAAsB;IACpC,OAAO,sBAAsB,CAAC,QAAQ,EAAE,CAAC;AAC3C,CAAC"}

package/dist/llm/stream-progress.d.ts

@@ -1,4 +1,3 @@
 export declare function markStreamEmittedBytes<E>(error: E, bytes: number): E;
 export declare function streamEmittedBytes(error: unknown): number;
-/** True when the failed attempt already delivered bytes to the caller. */
 export declare function streamAlreadyEmitted(error: unknown): boolean;

package/dist/llm/stream-progress.js

@@ -1,16 +1,3 @@
-/**
- * Stream progress tagging.
- *
- * A retry is only transparent if nothing has been emitted yet. Once tokens have
- * reached the caller's `onToken` sink (which the agent runner feeds into a
- * single delta parser and a single assistant message), re-running the request
- * appends a second copy of the answer — duplicated prose and, worse, duplicated
- * or spliced tool-call JSON.
- *
- * Errors thrown after emission are tagged with the byte count so every retry
- * decision (key rotation, provider fallback, the runner's recovery ladder) can
- * refuse to retry transparently.
- */
 const EMITTED_BYTES = Symbol.for("clai.stream.emittedBytes");
 export function markStreamEmittedBytes(error, bytes) {
     if (bytes > 0 && typeof error === "object" && error !== null) {
@@ -23,7 +10,6 @@ export function markStreamEmittedBytes(error, bytes) {
             });
         }
         catch {
-            // Frozen error objects cannot be tagged; treat as no progress.
         }
     }
     return error;
@@ -36,7 +22,6 @@ export function streamEmittedBytes(error) {
         ? value
         : 0;
 }
-/** True when the failed attempt already delivered bytes to the caller. */
 export function streamAlreadyEmitted(error) {
     return streamEmittedBytes(error) > 0;
 }

package/dist/llm/stream-progress.js.map

@@ -1 +1 @@
-{"version":3,"file":"stream-progress.js","sourceRoot":"","sources":["../../src/llm/stream-progress.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AACH,MAAM,aAAa,GAAG,MAAM,CAAC,GAAG,CAAC,0BAA0B,CAAC,CAAC;AAE7D,MAAM,UAAU,sBAAsB,CAAI,KAAQ,EAAE,KAAa;IAC/D,IAAI,KAAK,GAAG,CAAC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;QAC7D,IAAI,CAAC;YACH,MAAM,CAAC,cAAc,CAAC,KAAK,EAAE,aAAa,EAAE;gBAC1C,KAAK,EAAE,KAAK;gBACZ,UAAU,EAAE,KAAK;gBACjB,YAAY,EAAE,IAAI;gBAClB,QAAQ,EAAE,IAAI;aACf,CAAC,CAAC;QACL,CAAC;QAAC,MAAM,CAAC;YACP,+DAA+D;QACjE,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,KAAc;IAC/C,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,CAAC,CAAC;IAC1D,MAAM,KAAK,GAAI,KAAiC,CAAC,aAAa,CAAC,CAAC;IAChE,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC;QACrE,CAAC,CAAC,KAAK;QACP,CAAC,CAAC,CAAC,CAAC;AACR,CAAC;AAED,0EAA0E;AAC1E,MAAM,UAAU,oBAAoB,CAAC,KAAc;IACjD,OAAO,kBAAkB,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;AACvC,CAAC"}
+{"version":3,"file":"stream-progress.js","sourceRoot":"","sources":["../../src/llm/stream-progress.ts"],"names":[],"mappings":"AAAA,MAAM,aAAa,GAAG,MAAM,CAAC,GAAG,CAAC,0BAA0B,CAAC,CAAC;AAE7D,MAAM,UAAU,sBAAsB,CAAI,KAAQ,EAAE,KAAa;IAC/D,IAAI,KAAK,GAAG,CAAC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;QAC7D,IAAI,CAAC;YACH,MAAM,CAAC,cAAc,CAAC,KAAK,EAAE,aAAa,EAAE;gBAC1C,KAAK,EAAE,KAAK;gBACZ,UAAU,EAAE,KAAK;gBACjB,YAAY,EAAE,IAAI;gBAClB,QAAQ,EAAE,IAAI;aACf,CAAC,CAAC;QACL,CAAC;QAAC,MAAM,CAAC;QACT,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,KAAc;IAC/C,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,CAAC,CAAC;IAC1D,MAAM,KAAK,GAAI,KAAiC,CAAC,aAAa,CAAC,CAAC;IAChE,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC;QACrE,CAAC,CAAC,KAAK;QACP,CAAC,CAAC,CAAC,CAAC;AACR,CAAC;AAED,MAAM,UAAU,oBAAoB,CAAC,KAAc;IACjD,OAAO,kBAAkB,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;AACvC,CAAC"}

package/dist/llm/system-messages.d.ts

@@ -1,36 +1,14 @@
 import type { ChatMessage } from "../types.js";
-/**
- * Marker prefix used when a mid-conversation `system` message has to be
- * delivered as a user turn because the provider dialect only has a single
- * top-level system field (Anthropic `system`, Gemini `systemInstruction`).
- */
 export declare const SYSTEM_TURN_MARKER = "[SYSTEM]";
-/** Mutable, current-turn authority promoted into provider system fields. */
 export declare const REQUEST_CONTEXT_PREFIX = "REQUEST CONTEXT";
 export declare function isRequestContextSystemMessage(message: ChatMessage): boolean;
 export declare function requestContextSystemPrompts(messages: readonly ChatMessage[]): string[];
 export declare function withoutRequestContextSystemMessages(messages: readonly ChatMessage[]): ChatMessage[];
 export declare function upsertRequestContextMessage(messages: ChatMessage[], content: string): void;
-/**
- * One shared ordered normalization contract for dialects with a
- * single system slot.
- *
- * The first system message becomes the dialect's system field. Every later
- * system message (compaction memory, live plan, engagement scope, Responder
- * ledger, loop-guard/progress steering) is preserved *in place* as a user turn
- * tagged with {@link SYSTEM_TURN_MARKER}, so ordering relative to tool groups
- * is untouched and nothing is silently dropped.
- *
- * The marker is applied exactly once: content that already carries it is
- * passed through unchanged.
- */
 export declare function normalizeSystemMessages(messages: ChatMessage[]): {
-    /** Content of the first system message, if any. */
     systemPrompt: string | undefined;
-    /** History with later system messages converted to marked user turns. */
     rest: ChatMessage[];
 };
 export declare function singleLeadingSystemMessages(messages: readonly ChatMessage[]): ChatMessage[];
 export declare function markSystemTurn(content: string): string;
-/** Content of the first system message, matching the dialect system field. */
 export declare function firstSystemPrompt(messages: ChatMessage[]): string | undefined;

package/dist/llm/system-messages.js

@@ -1,10 +1,4 @@
-/**
- * Marker prefix used when a mid-conversation `system` message has to be
- * delivered as a user turn because the provider dialect only has a single
- * top-level system field (Anthropic `system`, Gemini `systemInstruction`).
- */
 export const SYSTEM_TURN_MARKER = "[SYSTEM]";
-/** Mutable, current-turn authority promoted into provider system fields. */
 export const REQUEST_CONTEXT_PREFIX = "REQUEST CONTEXT";
 export function isRequestContextSystemMessage(message) {
     return (message.role === "system" &&
@@ -20,24 +14,15 @@ export function withoutRequestContextSystemMessages(messages) {
 }
 export function upsertRequestContextMessage(messages, content) {
     for (let index = messages.length - 1; index >= 0; index -= 1) {
-        if (isRequestContextSystemMessage(messages[index]))
-            messages.splice(index, 1);
+        const message = messages[index];
+        if (isRequestContextSystemMessage(message)) {
+            if (message.content === content)
+                return;
+            break;
+        }
     }
     messages.push({ role: "system", content });
 }
-/**
- * One shared ordered normalization contract for dialects with a
- * single system slot.
- *
- * The first system message becomes the dialect's system field. Every later
- * system message (compaction memory, live plan, engagement scope, Responder
- * ledger, loop-guard/progress steering) is preserved *in place* as a user turn
- * tagged with {@link SYSTEM_TURN_MARKER}, so ordering relative to tool groups
- * is untouched and nothing is silently dropped.
- *
- * The marker is applied exactly once: content that already carries it is
- * passed through unchanged.
- */
 export function normalizeSystemMessages(messages) {
     let systemPrompt;
     let seenFirstSystem = false;
@@ -71,7 +56,6 @@ export function markSystemTurn(content) {
         ? content
         : `${SYSTEM_TURN_MARKER}\n${content}`;
 }
-/** Content of the first system message, matching the dialect system field. */
 export function firstSystemPrompt(messages) {
     return messages.find((message) => message.role === "system")?.content;
 }

package/dist/llm/system-messages.js.map

@@ -1 +1 @@
-{"version":3,"file":"system-messages.js","sourceRoot":"","sources":["../../src/llm/system-messages.ts"],"names":[],"mappings":"AAEA;;;;GAIG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAG,UAAU,CAAC;AAE7C,4EAA4E;AAC5E,MAAM,CAAC,MAAM,sBAAsB,GAAG,iBAAiB,CAAC;AAExD,MAAM,UAAU,6BAA6B,CAAC,OAAoB;IAChE,OAAO,CACL,OAAO,CAAC,IAAI,KAAK,QAAQ;QACzB,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,sBAAsB,CAAC,CACnD,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,2BAA2B,CACzC,QAAgC;IAEhC,OAAO,QAAQ;SACZ,MAAM,CAAC,6BAA6B,CAAC;SACrC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;AACvC,CAAC;AAED,MAAM,UAAU,mCAAmC,CACjD,QAAgC;IAEhC,OAAO,QAAQ,CAAC,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,6BAA6B,CAAC,OAAO,CAAC,CAAC,CAAC;AAC/E,CAAC;AAED,MAAM,UAAU,2BAA2B,CACzC,QAAuB,EACvB,OAAe;IAEf,KAAK,IAAI,KAAK,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAC7D,IAAI,6BAA6B,CAAC,QAAQ,CAAC,KAAK,CAAE,CAAC;YAAE,QAAQ,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACjF,CAAC;IACD,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC;AAC7C,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,uBAAuB,CAAC,QAAuB;IAM7D,IAAI,YAAgC,CAAC;IACrC,IAAI,eAAe,GAAG,KAAK,CAAC;IAC5B,MAAM,IAAI,GAAkB,EAAE,CAAC;IAC/B,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,IAAI,OAAO,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YAC9B,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YACnB,SAAS;QACX,CAAC;QACD,IAAI,CAAC,eAAe,EAAE,CAAC;YACrB,eAAe,GAAG,IAAI,CAAC;YACvB,YAAY,GAAG,OAAO,CAAC,OAAO,CAAC;YAC/B,SAAS;QACX,CAAC;QACD,IAAI,CAAC,IAAI,CAAC;YACR,GAAG,OAAO;YACV,IAAI,EAAE,MAAM;YACZ,OAAO,EAAE,cAAc,CAAC,OAAO,CAAC,OAAO,CAAC;SACzC,CAAC,CAAC;IACL,CAAC;IACD,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,2BAA2B,CACzC,QAAgC;IAEhC,MAAM,EAAE,YAAY,EAAE,IAAI,EAAE,GAAG,uBAAuB,CAAC,CAAC,GAAG,QAAQ,CAAC,CAAC,CAAC;IACtE,OAAO,YAAY,KAAK,SAAS;QAC/B,CAAC,CAAC,IAAI;QACN,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,YAAY,EAAE,EAAE,GAAG,IAAI,CAAC,CAAC;AAC3D,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,OAAe;IAC5C,OAAO,OAAO,CAAC,UAAU,CAAC,kBAAkB,CAAC;QAC3C,CAAC,CAAC,OAAO;QACT,CAAC,CAAC,GAAG,kBAAkB,KAAK,OAAO,EAAE,CAAC;AAC1C,CAAC;AAED,8EAA8E;AAC9E,MAAM,UAAU,iBAAiB,CAC/B,QAAuB;IAEvB,OAAO,QAAQ,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,KAAK,QAAQ,CAAC,EAAE,OAAO,CAAC;AACxE,CAAC"}
+{"version":3,"file":"system-messages.js","sourceRoot":"","sources":["../../src/llm/system-messages.ts"],"names":[],"mappings":"AAEA,MAAM,CAAC,MAAM,kBAAkB,GAAG,UAAU,CAAC;AAE7C,MAAM,CAAC,MAAM,sBAAsB,GAAG,iBAAiB,CAAC;AAExD,MAAM,UAAU,6BAA6B,CAAC,OAAoB;IAChE,OAAO,CACL,OAAO,CAAC,IAAI,KAAK,QAAQ;QACzB,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,sBAAsB,CAAC,CACnD,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,2BAA2B,CACzC,QAAgC;IAEhC,OAAO,QAAQ;SACZ,MAAM,CAAC,6BAA6B,CAAC;SACrC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;AACvC,CAAC;AAED,MAAM,UAAU,mCAAmC,CACjD,QAAgC;IAEhC,OAAO,QAAQ,CAAC,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,6BAA6B,CAAC,OAAO,CAAC,CAAC,CAAC;AAC/E,CAAC;AAED,MAAM,UAAU,2BAA2B,CACzC,QAAuB,EACvB,OAAe;IAEf,KAAK,IAAI,KAAK,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAC7D,MAAM,OAAO,GAAG,QAAQ,CAAC,KAAK,CAAE,CAAC;QACjC,IAAI,6BAA6B,CAAC,OAAO,CAAC,EAAE,CAAC;YAC3C,IAAI,OAAO,CAAC,OAAO,KAAK,OAAO;gBAAE,OAAO;YACxC,MAAM;QACR,CAAC;IACH,CAAC;IACD,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC;AAC7C,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,QAAuB;IAI7D,IAAI,YAAgC,CAAC;IACrC,IAAI,eAAe,GAAG,KAAK,CAAC;IAC5B,MAAM,IAAI,GAAkB,EAAE,CAAC;IAC/B,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,IAAI,OAAO,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YAC9B,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YACnB,SAAS;QACX,CAAC;QACD,IAAI,CAAC,eAAe,EAAE,CAAC;YACrB,eAAe,GAAG,IAAI,CAAC;YACvB,YAAY,GAAG,OAAO,CAAC,OAAO,CAAC;YAC/B,SAAS;QACX,CAAC;QACD,IAAI,CAAC,IAAI,CAAC;YACR,GAAG,OAAO;YACV,IAAI,EAAE,MAAM;YACZ,OAAO,EAAE,cAAc,CAAC,OAAO,CAAC,OAAO,CAAC;SACzC,CAAC,CAAC;IACL,CAAC;IACD,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,2BAA2B,CACzC,QAAgC;IAEhC,MAAM,EAAE,YAAY,EAAE,IAAI,EAAE,GAAG,uBAAuB,CAAC,CAAC,GAAG,QAAQ,CAAC,CAAC,CAAC;IACtE,OAAO,YAAY,KAAK,SAAS;QAC/B,CAAC,CAAC,IAAI;QACN,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,YAAY,EAAE,EAAE,GAAG,IAAI,CAAC,CAAC;AAC3D,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,OAAe;IAC5C,OAAO,OAAO,CAAC,UAAU,CAAC,kBAAkB,CAAC;QAC3C,CAAC,CAAC,OAAO;QACT,CAAC,CAAC,GAAG,kBAAkB,KAAK,OAAO,EAAE,CAAC;AAC1C,CAAC;AAED,MAAM,UAAU,iBAAiB,CAC/B,QAAuB;IAEvB,OAAO,QAAQ,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,KAAK,QAAQ,CAAC,EAAE,OAAO,CAAC;AACxE,CAAC"}

package/dist/llm/token-estimate-calibration.d.ts

@@ -3,11 +3,6 @@ export interface RequestTokenCalibration {
     readonly ratio: number;
     readonly samples: number;
 }
-/**
- * Pair the estimate made for one request with the prompt size the provider
- * reported for it. Callers must pass the uncalibrated estimate, otherwise the
- * correction compounds against itself.
- */
 export declare function recordRequestTokenObservation(input: {
     readonly provider: ProviderId | undefined;
     readonly model: string | undefined;
@@ -15,7 +10,6 @@ export declare function recordRequestTokenObservation(input: {
     readonly actualPromptTokens: number;
 }): void;
 export declare function requestTokenCalibration(provider: ProviderId | undefined, model: string | undefined): RequestTokenCalibration | undefined;
-/** Apply the learned factor, or return the estimate unchanged when untrained. */
 export declare function calibratedRequestTokens(provider: ProviderId | undefined, model: string | undefined, estimatedTokens: number): number;
 export declare function resetRequestTokenCalibration(options?: {
     removePersisted?: boolean;