@voltro/ai 0.74.0 → 0.76.0

package/dist/inferenceQueue-t5ayqzjs.js

@@ -0,0 +1,1053 @@
+import { c as e, o as t, r as n } from "./provider-BsyDya-H.js";
+import { NoObjectGeneratedError as r, Output as i, experimental_transcribe as a, generateObject as o, generateText as s, jsonSchema as c, stepCountIs as l, tool as u, wrapLanguageModel as d } from "ai";
+import { Context as f, Data as p, Effect as m, JSONSchema as h, Metric as g, MetricBoundaries as ee, Runtime as te, Schema as _, Stream as ne } from "effect";
+import { and as v, bigint as re, eq as y, generateId as b, gte as ie, id as x, integer as S, isNull as ae, json as C, queryFor as w, registerRetention as T, retentionTtlMsFromEnv as oe, table as E, text as D, timestamp as O } from "@voltro/database";
+import { Ajv as se } from "ajv";
+import { Ajv2019 as ce } from "ajv/dist/2019.js";
+import { Ajv2020 as le } from "ajv/dist/2020.js";
+import ue from "ajv-formats";
+import { createHash as de } from "node:crypto";
+//#region src/tools.ts
+var fe = (e, t) => e(t), pe = Symbol.for("@voltro/ai/Tool"), me = (e) => ({
+	...e,
+	[pe]: !0
+}), he = (e) => typeof e == "object" && !!e && e[pe] === !0, ge = (e) => {
+	let t = e.default;
+	if (typeof t != "function") return 0;
+	let n = 0;
+	for (let r of Object.values(e)) he(r) && r.execute === void 0 && r.handler === void 0 && (r.handler = t, n += 1);
+	return n;
+}, _e = (e, t, n) => {
+	let r = t ?? te.defaultRuntime;
+	return te.runPromise(r)(e, n === void 0 ? {} : { signal: n });
+}, ve = (e, t, n) => n ? e.pipe(m.timeoutFail({
+	duration: `${n} millis`,
+	onTimeout: () => /* @__PURE__ */ Error(`tool '${t}' timed out`)
+})) : e, ye = (e, t, n) => {
+	if (e.execute) return _e(ve(fe(e.execute, t), e.name, e.timeoutMs), n.runtime, n.abortSignal);
+	if (e.handler) {
+		let r = e.handler, i = r(t, n);
+		return m.isEffect(i) ? _e(ve(i, e.name, e.timeoutMs), n.runtime, n.abortSignal) : i;
+	}
+	return Promise.reject(/* @__PURE__ */ Error(`tool '${e.name}' has neither execute nor handler`));
+}, be = (e) => {
+	let t = h.make(e);
+	return t && t.type === "object" ? t : {
+		type: "object",
+		properties: {},
+		additionalProperties: !1
+	};
+}, xe = (e) => {
+	let t = e.inputJsonSchema;
+	return t === void 0 ? be(e.input) : t.type === "object" ? t : {
+		type: "object",
+		properties: {},
+		additionalProperties: !0
+	};
+}, k = (e, t = { store: void 0 }) => {
+	let n = {};
+	for (let [r, i] of Object.entries(e)) n[i.name || r] = u({
+		description: i.description,
+		inputSchema: c(xe(i)),
+		execute: async (e, n) => {
+			let r = t.abortSignal, a = n?.abortSignal, o = r !== void 0 && a !== void 0 && r !== a ? AbortSignal.any([r, a]) : a ?? r;
+			o?.throwIfAborted();
+			let s = {
+				...t,
+				...n?.toolCallId === void 0 ? {} : { invocationId: n.toolCallId },
+				...o === void 0 ? {} : { abortSignal: o }
+			}, c = await m.runPromise(_.decodeUnknown(i.input)(e));
+			o?.throwIfAborted();
+			let l = await ye(i, c, s);
+			return i.output ? m.runPromise(_.decodeUnknown(i.output)(l)) : l;
+		}
+	});
+	return n;
+}, A = class extends _.TaggedError()("AiBudgetExceeded", {
+	tenantId: _.NullOr(_.String),
+	limitUsd: _.Number,
+	spentUsd: _.Number,
+	attemptedUsd: _.Number
+}) {}, j = class extends _.TaggedError()("AiBudgetInvalidInput", {
+	field: _.Literal("limitUsd", "addUsd", "since", "reservationId", "tenantId"),
+	message: _.String
+}) {}, M = class extends _.TaggedError()("AiBudgetReservationConflict", {
+	reason: _.Literal("identity-bound", "already-released"),
+	message: _.String
+}) {}, N = "_voltro_ai_budget", P = "_voltro_ai_budget_reservations", F = E(N, {
+	id: x({ prefix: "aib" }),
+	bucket: D().maxLength(191),
+	tenantId: D().nullable(),
+	reservedMicroUsd: re().default("0").narrowedFrom("integer"),
+	updatedAt: O().default("now")
+}).unique("byAiBudgetBucket", ["bucket"]), Se = E(P, {
+	id: x({ prefix: "air" }),
+	reservationId: D().maxLength(191),
+	bucket: D().maxLength(191),
+	tenantId: D().nullable(),
+	amountMicroUsd: re(),
+	releasedAt: O().nullable(),
+	updatedAt: O().default("now")
+}).unique("byAiBudgetReservation", ["reservationId"]), I = class extends Error {}, Ce = (e, t, n) => {
+	if (!Number.isFinite(e) || e < 0) throw new j({
+		field: t,
+		message: `${t} must be finite and non-negative`
+	});
+	let r = e * 1e6, i = n === "up" ? Math.ceil(r) : Math.floor(r);
+	if (!Number.isSafeInteger(i)) throw new j({
+		field: t,
+		message: `${t} exceeds the safe micro-USD range`
+	});
+	return BigInt(i);
+}, L = (e) => {
+	if (typeof e == "number" && !Number.isSafeInteger(e)) throw Error("AI budget counter is not an exact integer");
+	let t = String(e);
+	if (!/^(0|[1-9][0-9]*)$/.test(t)) throw Error("AI budget counter is invalid; refusing to assume zero");
+	return BigInt(t);
+}, we = (e) => {
+	if (typeof e != "string" || e.length === 0 || e.length > 191) throw new j({
+		field: "reservationId",
+		message: "AI budget reservationId must contain 1–191 characters"
+	});
+	return e;
+}, Te = () => {
+	let e = oe(process.env.VOLTRO_AI_USAGE_TTL_HOURS, 8760);
+	for (let t of [N, P]) T({
+		source: "plugin",
+		table: t,
+		timeColumn: "updatedAt",
+		ttlMs: e
+	});
+}, Ee = async (e, t) => {
+	for (let n = 0;; n++) try {
+		return await e.transactional(t);
+	} catch (e) {
+		if (!(e instanceof I) || n >= 31) throw e;
+		await new Promise((e) => setTimeout(e, n + 1));
+	}
+}, De = (e) => e instanceof A || e instanceof j || e instanceof M ? m.fail(e) : m.die(e), Oe = (e) => m.tryPromise({
+	try: e,
+	catch: (e) => e
+}).pipe(m.catchAll(De)), ke = (e, t) => m.gen(function* () {
+	let { limit: n, amount: r, tenantId: i, bucket: a, reservationId: o } = yield* m.try({
+		try: () => {
+			let e = Ce(t.limitUsd, "limitUsd", "down"), n = Ce(t.addUsd ?? 0, "addUsd", "up"), r = t.tenantId ?? null;
+			if (t.since !== void 0 && (!(t.since instanceof Date) || !Number.isFinite(t.since.getTime()))) throw new j({
+				field: "since",
+				message: "AI budget since must be a valid Date"
+			});
+			let i = `${r ?? "global"}::${t.since === void 0 ? "all" : t.since.getTime()}`;
+			if (i.length > 191) throw new j({
+				field: "tenantId",
+				message: "AI budget tenant/window key exceeds 191 characters"
+			});
+			return {
+				limit: e,
+				amount: n,
+				tenantId: r,
+				bucket: i,
+				reservationId: we(t.reservationId ?? b({
+					kind: "typeid",
+					prefix: "air"
+				}, "_voltro_ai_budget_reservations"))
+			};
+		},
+		catch: (e) => e
+	}).pipe(m.catchAll(De));
+	Te();
+	let s = async (e) => {
+		let [t] = await e.query(w(F).where(y("bucket", a)).descriptor);
+		if (t && t.tenantId !== i) throw Error("AI budget bucket belongs to a different tenant");
+		return t;
+	}, c = (e) => {
+		if (r === 0n ? e >= n : e + r > n) throw new A({
+			tenantId: i,
+			limitUsd: t.limitUsd,
+			spentUsd: Number(e) / 1e6,
+			attemptedUsd: t.addUsd ?? 0
+		});
+	};
+	return r === 0n ? yield* Oe(async () => {
+		let t = await s(e);
+		return c(t ? L(t.reservedMicroUsd) : 0n), null;
+	}) : yield* Oe(() => Ee(e, async (e) => {
+		let t = b({
+			kind: "typeid",
+			prefix: "air"
+		}, P), n = /* @__PURE__ */ new Date(), l = await e.insertIgnore(P, {
+			id: t,
+			reservationId: o,
+			bucket: a,
+			tenantId: i,
+			amountMicroUsd: String(r),
+			releasedAt: null,
+			updatedAt: n
+		}, { conflictColumns: ["reservationId"] });
+		if (l.id !== t) {
+			if (l.tenantId !== i || l.bucket !== a || L(l.amountMicroUsd) !== r) throw new M({
+				reason: "identity-bound",
+				message: "AI budget reservationId is already bound to another tenant, amount or window"
+			});
+			if (l.releasedAt !== null) throw new M({
+				reason: "already-released",
+				message: "AI budget reservation was released; new spend requires a new reservationId"
+			});
+			return {
+				reservationId: o,
+				tenantId: i,
+				amountMicroUsd: String(r),
+				status: "replayed"
+			};
+		}
+		let u = await s(e), d = u ? L(u.reservedMicroUsd) : 0n;
+		if (c(d), u) {
+			if (await e.updateMany("_voltro_ai_budget", {
+				reservedMicroUsd: String(d + r),
+				updatedAt: n
+			}, { where: v(y("id", u.id), y("reservedMicroUsd", u.reservedMicroUsd)) }) !== 1) throw new I("AI budget counter changed; discard this receipt and retry");
+		} else {
+			let t = b({
+				kind: "typeid",
+				prefix: "aib"
+			}, N);
+			if ((await e.insertIgnore("_voltro_ai_budget", {
+				id: t,
+				bucket: a,
+				tenantId: i,
+				reservedMicroUsd: String(r),
+				updatedAt: n
+			}, { conflictColumns: ["bucket"] })).id !== t) throw new I("AI budget counter was created concurrently; discard this receipt and retry");
+		}
+		return {
+			reservationId: o,
+			tenantId: i,
+			amountMicroUsd: String(r),
+			status: "reserved"
+		};
+	}));
+}), Ae = (e, t) => m.gen(function* () {
+	return yield* m.try({
+		try: () => we(t.reservationId),
+		catch: (e) => e
+	}).pipe(m.catchAll((e) => e instanceof j ? m.fail(e) : m.die(e))), yield* m.promise(() => Ee(e, async (e) => {
+		let n = w(Se).where(v(y("reservationId", t.reservationId), t.tenantId === null ? ae("tenantId") : y("tenantId", t.tenantId))).descriptor, [r] = await e.query(n);
+		if (!r) return "not-found";
+		if (r.releasedAt !== null) return "already-released";
+		let i = /* @__PURE__ */ new Date();
+		if (await e.updateMany("_voltro_ai_budget_reservations", {
+			releasedAt: i,
+			updatedAt: i
+		}, { where: v(y("id", r.id), ae("releasedAt")) }) !== 1) {
+			let [t] = await e.query(n);
+			if (t !== void 0 && t.releasedAt !== null) return "already-released";
+			throw new I("AI budget release raced; retry");
+		}
+		let [a] = await e.query(w(F).where(y("bucket", r.bucket)).descriptor);
+		if (!a || a.tenantId !== t.tenantId) throw Error("AI budget counter missing or mismatched; refusing unproved credit");
+		let o = L(a.reservedMicroUsd), s = L(r.amountMicroUsd);
+		if (s > o) throw Error("AI budget receipt exceeds its counter; refusing unproved credit");
+		if (await e.updateMany("_voltro_ai_budget", {
+			reservedMicroUsd: String(o - s),
+			updatedAt: i
+		}, { where: v(y("id", a.id), y("reservedMicroUsd", a.reservedMicroUsd)) }) !== 1) throw new I("AI budget counter changed; discard this release and retry");
+		return "released";
+	}));
+}), je = {
+	"claude-opus-4-8": {
+		inputPer1M: 15,
+		outputPer1M: 75
+	},
+	"claude-opus-4-6": {
+		inputPer1M: 15,
+		outputPer1M: 75
+	},
+	"claude-sonnet-4-6": {
+		inputPer1M: 3,
+		outputPer1M: 15
+	},
+	"claude-haiku-4-5": {
+		inputPer1M: 1,
+		outputPer1M: 5
+	},
+	"claude-haiku-4-5-20251001": {
+		inputPer1M: 1,
+		outputPer1M: 5
+	},
+	"gpt-5.5": {
+		inputPer1M: 5,
+		outputPer1M: 30
+	},
+	"gpt-4o": {
+		inputPer1M: 2.5,
+		outputPer1M: 10
+	},
+	"gpt-4o-mini": {
+		inputPer1M: .15,
+		outputPer1M: .6
+	},
+	"gemini-2.5-pro": {
+		inputPer1M: 1.25,
+		outputPer1M: 10
+	},
+	"gemini-2.5-flash": {
+		inputPer1M: .3,
+		outputPer1M: 2.5
+	},
+	"gemini-2.0-flash": {
+		inputPer1M: .1,
+		outputPer1M: .4
+	}
+}, R = {}, Me = (e) => {
+	let t = R;
+	return R = { ...e }, t;
+}, Ne = () => ({
+	...je,
+	...R
+}), Pe = (e) => {
+	let t = Ne();
+	if (t[e] !== void 0) return t[e];
+	let n = e.indexOf("/");
+	if (n > 0) {
+		let r = e.slice(n + 1);
+		if (t[r] !== void 0) return t[r];
+	}
+}, z = (e) => typeof e.unit == "string", Fe = (e) => z(e) ? {
+	unit: e.unit,
+	input: e.input ?? 0,
+	output: e.output ?? 0
+} : {
+	unit: "tokens",
+	input: e.inputTokens ?? 0,
+	output: e.outputTokens ?? 0
+}, Ie = (e, t) => {
+	let { unit: n, input: r, output: i } = Fe(e);
+	if (t.actualCostUsd !== void 0) {
+		if (!Number.isFinite(t.actualCostUsd) || t.actualCostUsd < 0) throw RangeError("estimateCostUsd: actualCostUsd must be a finite, non-negative USD amount.");
+		let e = r + i, a = e > 0 ? r / e : 0;
+		return {
+			inputTokens: r,
+			outputTokens: i,
+			inputCostUsd: t.actualCostUsd * a,
+			outputCostUsd: t.actualCostUsd * (1 - a),
+			totalCostUsd: t.actualCostUsd,
+			costSource: "gateway",
+			unit: n
+		};
+	}
+	if (n !== "tokens" && t.price === void 0) return {
+		inputTokens: r,
+		outputTokens: i,
+		inputCostUsd: 0,
+		outputCostUsd: 0,
+		totalCostUsd: 0,
+		costSource: "unpriced",
+		unit: n
+	};
+	let a = t.price ?? Pe(t.model) ?? {
+		inputPer1M: 0,
+		outputPer1M: 0
+	}, o = r / 1e6 * a.inputPer1M, s = i / 1e6 * a.outputPer1M;
+	return {
+		inputTokens: r,
+		outputTokens: i,
+		inputCostUsd: o,
+		outputCostUsd: s,
+		totalCostUsd: o + s,
+		costSource: "estimated",
+		unit: n
+	};
+}, B = "_voltro_ai_usage", Le = E(B, {
+	id: x({ prefix: "aiu" }),
+	tenantId: D().nullable(),
+	provider: D(),
+	model: D(),
+	operation: D(),
+	agent: D().nullable(),
+	inputTokens: S().default(0),
+	outputTokens: S().default(0),
+	costMicroUsd: S().default(0),
+	costSource: D().default("estimated"),
+	unit: D().default("tokens"),
+	promptId: D().nullable(),
+	promptDigest: D().nullable(),
+	promptRevision: S().nullable(),
+	requestId: D().maxLength(191).nullable(),
+	attempt: S().nullable(),
+	receiptKey: D().maxLength(191).backfill((e) => `aiu-pre-receipt-${String(e.id)}`),
+	calledAt: O().default("now")
+}).index(["tenantId"]).index(["promptDigest"]).unique("byAiUsageReceipt", ["receiptKey"]), Re = (e) => {
+	let t = e?.gateway?.cost, n = typeof t == "string" ? t.trim() : void 0, r = typeof t == "number" ? t : n !== void 0 && /^(?:\d+(?:\.\d*)?|\.\d+)(?:e[+-]?\d+)?$/i.test(n) ? Number(n) : NaN;
+	return Number.isFinite(r) && r >= 0 ? r : void 0;
+}, ze = class extends p.TaggedError("AiUsageReceiptConflict") {
+	get message() {
+		let e = Object.keys(this.recorded).filter((e) => this.recorded[e] !== this.attempted[e]).map((e) => `${e} ${String(this.recorded[e])} → ${String(this.attempted[e])}`);
+		return `AI usage receipt ${this.receiptKey} was recorded again with different values (${e.join(", ")}). A replay of the same paid call has the same values; the row keeps what was recorded first.`;
+	}
+}, Be = class extends p.TaggedError("AiUsageLedgerError") {
+	get message() {
+		return `AI usage receipt ${this.receiptKey} could not be written: ${this.cause instanceof Error ? this.cause.message : String(this.cause)}`;
+	}
+}, Ve = (e) => ({
+	tenantId: e.tenantId ?? null,
+	provider: String(e.provider),
+	model: String(e.model),
+	operation: String(e.operation),
+	inputTokens: Number(e.inputTokens),
+	outputTokens: Number(e.outputTokens),
+	costMicroUsd: Number(e.costMicroUsd)
+}), He = (e, t) => (T({
+	source: "plugin",
+	table: B,
+	timeColumn: "calledAt",
+	ttlMs: oe(process.env.VOLTRO_AI_USAGE_TTL_HOURS, 8760)
+}), m.gen(function* () {
+	let n = Ie(t.usage, {
+		model: t.model,
+		...t.price === void 0 ? {} : { price: t.price },
+		...t.actualCostUsd === void 0 ? {} : { actualCostUsd: t.actualCostUsd }
+	}), r = t.receiptKey ?? (t.requestId === void 0 ? b({
+		kind: "typeid",
+		prefix: "aiu"
+	}, "_voltro_ai_usage") : `${t.requestId}#${t.attempt ?? 0}`), i = {
+		tenantId: t.tenantId ?? null,
+		provider: t.provider,
+		model: t.model,
+		operation: t.operation,
+		agent: t.agent ?? null,
+		inputTokens: n.inputTokens,
+		outputTokens: n.outputTokens,
+		costMicroUsd: Math.round(n.totalCostUsd * 1e6),
+		costSource: n.costSource,
+		unit: n.unit,
+		promptId: t.prompt?.promptId ?? null,
+		promptDigest: t.prompt?.digest ?? null,
+		promptRevision: t.prompt?.revision ?? null,
+		requestId: t.requestId ?? null,
+		attempt: t.attempt ?? null,
+		receiptKey: r,
+		calledAt: t.calledAt ?? /* @__PURE__ */ new Date()
+	}, a = yield* m.tryPromise({
+		try: () => e.insertIgnoreWithOutcome(B, i, { conflictColumns: ["receiptKey"] }),
+		catch: (e) => new Be({
+			receiptKey: String(r),
+			cause: e
+		})
+	});
+	if (a.outcome === "inserted") return {
+		...n,
+		receipt: "recorded"
+	};
+	let o = Ve(a.row), s = Ve(i);
+	return Object.keys(o).some((e) => o[e] !== s[e]) ? yield* m.fail(new ze({
+		receiptKey: String(r),
+		recorded: o,
+		attempted: s
+	})) : {
+		...Ue(a.row, n),
+		receipt: "replayed"
+	};
+})), Ue = (e, t) => {
+	let n = Number(e.costMicroUsd) / 1e6;
+	return {
+		...t,
+		inputTokens: Number(e.inputTokens),
+		outputTokens: Number(e.outputTokens),
+		totalCostUsd: n,
+		costSource: e.costSource ?? t.costSource
+	};
+}, We = (e, t) => He(e, t).pipe(m.asVoid, m.catchAllCause((e) => m.logError("AI usage NOT booked as recorded — the call itself succeeded and was returned", e).pipe(m.annotateLogs({
+	scope: "voltro:ai:usage",
+	provider: t.provider,
+	model: t.model,
+	operation: t.operation,
+	...z(t.usage) ? {
+		unit: t.usage.unit,
+		input: t.usage.input ?? null,
+		output: t.usage.output ?? null
+	} : {
+		inputTokens: t.usage.inputTokens ?? null,
+		outputTokens: t.usage.outputTokens ?? null
+	},
+	...t.requestId === void 0 ? {} : {
+		requestId: t.requestId,
+		attempt: t.attempt ?? 0
+	},
+	...t.actualCostUsd === void 0 ? {} : { actualCostUsd: t.actualCostUsd }
+})))), Ge = (e, t = {}) => m.promise(async () => {
+	let n = [
+		...t.tenantId === void 0 ? [] : [y("tenantId", t.tenantId)],
+		...t.since === void 0 ? [] : [ie("calledAt", t.since)],
+		...t.promptDigest === void 0 ? [] : [y("promptDigest", t.promptDigest)]
+	], r = w(Le), i = n.length === 0 ? r : r.where(n.length === 1 ? n[0] : v(...n));
+	return (await e.query(i.descriptor)).reduce((e, t) => e + (Number(t.costMicroUsd) || 0), 0) / 1e6;
+}), Ke = ee.exponential({
+	start: .001,
+	factor: 2,
+	count: 16
+}), V = g.counter("voltro_ai_calls_total", { description: "Total AI provider calls (generate + stream)." }), qe = g.counter("voltro_ai_call_errors_total", { description: "AI provider calls that errored." }), H = g.histogram("voltro_ai_call_duration_seconds", Ke, "AI provider call duration in seconds."), Je = g.counter("voltro_ai_input_tokens_total", {
+	description: "Total prompt (input) tokens across AI calls.",
+	incremental: !0
+}), Ye = g.counter("voltro_ai_output_tokens_total", {
+	description: "Total completion (output) tokens across AI calls.",
+	incremental: !0
+}), Xe = g.counter("voltro_ai_cost_microusd_total", {
+	description: "Total estimated AI spend in micro-USD (USD × 1e6).",
+	incremental: !0
+}), U = (e, t) => ({
+	provider: e.name,
+	model: e.model ?? e.name,
+	operation: t
+}), W = (e, t, n) => e.pipe(m.tagMetrics("provider", t.provider), m.tagMetrics("model", t.model), m.tagMetrics("operation", t.operation), m.tagMetrics("status", n)), Ze = (e, t) => {
+	let n = t.inputTokens ?? 0, r = t.outputTokens ?? 0, i = Ie({
+		inputTokens: n,
+		outputTokens: r
+	}, { model: e.model }), a = Math.round(i.totalCostUsd * 1e6);
+	return m.all([
+		W(g.incrementBy(Je, n), e, "ok"),
+		W(g.incrementBy(Ye, r), e, "ok"),
+		W(g.incrementBy(Xe, a), e, "ok")
+	], { discard: !0 });
+}, G = (e, t, n) => m.gen(function* () {
+	let r = yield* m.clockWith((e) => e.currentTimeMillis);
+	return yield* n.pipe(m.tap((n) => Ze(e, t(n))), m.tap(() => W(g.increment(V), e, "ok").pipe(m.zipRight(m.clockWith((e) => e.currentTimeMillis).pipe(m.flatMap((t) => W(g.update(H, (t - r) / 1e3), e, "ok")))))), m.tapError(() => W(g.increment(V), e, "error").pipe(m.zipRight(W(g.increment(qe), e, "error")), m.zipRight(m.clockWith((e) => e.currentTimeMillis).pipe(m.flatMap((t) => W(g.update(H, (t - r) / 1e3), e, "error")))))), m.withSpan("voltro.ai.call", { attributes: {
+		"ai.provider": e.provider,
+		"ai.model": e.model,
+		"ai.operation": e.operation
+	} }));
+}), Qe = (e, t) => {
+	let n = Date.now();
+	return t.pipe(ne.tap((t) => t._tag === "done" ? W(g.increment(V), e, "ok").pipe(m.zipRight(W(g.update(H, (Date.now() - n) / 1e3), e, "ok")), m.zipRight(Ze(e, t.usage ?? {}))) : t._tag === "error" ? W(g.increment(V), e, "error").pipe(m.zipRight(W(g.increment(qe), e, "error")), m.zipRight(W(g.update(H, (Date.now() - n) / 1e3), e, "error"))) : m.void), ne.withSpan("voltro.ai.call", { attributes: {
+		"ai.provider": e.provider,
+		"ai.model": e.model,
+		"ai.operation": e.operation
+	} }));
+}, $e = (e) => {
+	let t = structuredClone(e), n = {
+		strict: !0,
+		strictTypes: !1,
+		strictTuples: !1,
+		strictRequired: !1,
+		allowUnionTypes: !0
+	}, r = t.$schema, i = r === "https://json-schema.org/draft/2020-12/schema" ? new le(n) : r === "https://json-schema.org/draft/2019-09/schema" ? new ce(n) : new se(n);
+	ue(i);
+	let a = i.compile(t);
+	if ("$async" in a && a.$async === !0) throw Error("asynchronous output schemas are not supported");
+	return c(t, { validate: (e) => a(e) ? {
+		success: !0,
+		value: e
+	} : {
+		success: !1,
+		error: /* @__PURE__ */ Error("model output does not satisfy the declared JSON Schema")
+	} });
+}, et = _.Number.pipe(_.int(), _.nonNegative(), _.lessThanOrEqualTo(2 ** 53 - 1)), tt = _.Struct({
+	provider: _.String,
+	model: _.String,
+	usage: _.Struct({
+		inputTokens: _.optional(et),
+		outputTokens: _.optional(et),
+		totalTokens: _.optional(et),
+		completeness: _.Literal("complete", "partial", "unknown")
+	}),
+	cost: _.Union(_.Struct({
+		status: _.Literal("actual"),
+		source: _.Literal("gateway"),
+		usd: _.Number.pipe(_.finite(), _.nonNegative())
+	}), _.Struct({
+		status: _.Literal("unknown"),
+		source: _.Literal("unreported")
+	}))
+}), nt = (e) => typeof e == "number" && Number.isSafeInteger(e) && e >= 0 ? e : void 0, K = (e, t, n, r) => {
+	let i = typeof n == "object" && n ? n : {}, a = nt("inputTokens" in i ? i.inputTokens : void 0), o = nt("outputTokens" in i ? i.outputTokens : void 0), s = nt("totalTokens" in i ? i.totalTokens : void 0), c = [
+		a,
+		o,
+		s
+	].filter((e) => e !== void 0).length, l = Re(r);
+	return {
+		provider: e,
+		model: t,
+		usage: {
+			...a === void 0 ? {} : { inputTokens: a },
+			...o === void 0 ? {} : { outputTokens: o },
+			...s === void 0 ? {} : { totalTokens: s },
+			completeness: c === 3 ? "complete" : c === 0 ? "unknown" : "partial"
+		},
+		cost: l === void 0 ? {
+			status: "unknown",
+			source: "unreported"
+		} : {
+			status: "actual",
+			source: "gateway",
+			usd: l
+		}
+	};
+}, rt = (e, t) => {
+	let n = [];
+	return {
+		model: d({
+			model: t,
+			middleware: {
+				specificationVersion: "v4",
+				wrapGenerate: async ({ doGenerate: r }) => {
+					let i = await r(), a = i.usage.inputTokens.total, o = i.usage.outputTokens.total;
+					return n.push(K(e, t.modelId, {
+						inputTokens: a,
+						outputTokens: o,
+						totalTokens: a !== void 0 && o !== void 0 ? a + o : void 0
+					}, i.providerMetadata)), i;
+				}
+			}
+		}),
+		evidence: (r) => {
+			if (n.length === 0) return K(e, t.modelId, r);
+			let i = (e) => {
+				let t = 0;
+				for (let r of n) {
+					let n = r.usage[e];
+					if (n === void 0) return;
+					t += n;
+				}
+				return t;
+			}, a = 0;
+			for (let e of n) {
+				if (e.cost.status !== "actual") {
+					a = void 0;
+					break;
+				}
+				a += e.cost.usd;
+			}
+			return K(e, t.modelId, {
+				inputTokens: i("inputTokens"),
+				outputTokens: i("outputTokens"),
+				totalTokens: i("totalTokens")
+			}, a === void 0 ? void 0 : { gateway: { cost: a } });
+		}
+	};
+}, q = class extends _.TaggedError()("AiError", {
+	reason: _.Literal("generation", "decode"),
+	message: _.String,
+	evidence: _.optional(tt)
+}) {}, J = (e, t = {}) => {
+	let n = String(e?.message ?? e), r = it(e), i = new q({
+		reason: "generation",
+		message: `${t.prefix === void 0 ? "" : `${t.prefix}: `}${n}${r.length > 0 ? ` (${r.join(" <- ")})` : ""}`,
+		...t.evidence === void 0 ? {} : { evidence: t.evidence }
+	});
+	return Object.defineProperty(i, "cause", {
+		value: e,
+		enumerable: !1,
+		configurable: !0,
+		writable: !0
+	}), i;
+}, it = (e) => {
+	let t = [], n = e;
+	for (let e = 0; e < 8 && typeof n == "object" && n; e++) {
+		let e = n;
+		typeof e.code == "string" && e.code.length > 0 && !t.includes(e.code) && t.push(e.code), n = typeof e.lastError == "object" && e.lastError !== null ? e.lastError : e.cause;
+	}
+	return t;
+}, at = (e, t, n) => {
+	let r = [e, ...t ?? []], i = (e) => n(r[e]).pipe(m.catchTag("AiError", (t) => t.reason === "generation" && e + 1 < r.length ? i(e + 1) : m.fail(t)));
+	return i(0);
+}, ot = (e) => m.gen(function* () {
+	let r = e.messages === void 0 ? e.prompt === void 0 ? void 0 : { prompt: e.prompt } : { messages: [...e.messages] };
+	if (r === void 0 || e.prompt !== void 0 && e.messages !== void 0) return yield* m.fail(new q({
+		reason: "generation",
+		message: "generateText: supply exactly one of prompt or messages"
+	}));
+	let i = e.tools === void 0 ? void 0 : yield* m.runtime(), a = e.tools === void 0 || i === void 0 ? void 0 : k(e.tools, {
+		store: void 0,
+		...e.toolContext,
+		...e.abortSignal === void 0 ? {} : { abortSignal: e.abortSignal },
+		runtime: i
+	});
+	return yield* at(e.provider ?? n(), e.fallbacks, (n) => G(U(n, "generateText"), (e) => e.usage, m.tryPromise({
+		try: async () => {
+			let i = t(n), o = await s({
+				model: i,
+				...r,
+				...e.abortSignal === void 0 ? {} : { abortSignal: e.abortSignal },
+				...a === void 0 ? {} : {
+					tools: a,
+					stopWhen: l(e.maxSteps ?? 8)
+				},
+				...e.system === void 0 ? {} : { system: e.system },
+				...e.maxTokens === void 0 ? {} : { maxOutputTokens: e.maxTokens },
+				...e.temperature === void 0 ? {} : { temperature: e.temperature },
+				...e.providerOptions === void 0 ? {} : { providerOptions: e.providerOptions }
+			}), c = a !== void 0 && o.totalUsage !== void 0 ? o.totalUsage : o.usage;
+			return {
+				text: o.text,
+				sources: o.sources,
+				usage: {
+					inputTokens: c.inputTokens,
+					outputTokens: c.outputTokens,
+					totalTokens: c.totalTokens
+				},
+				providerMetadata: o.providerMetadata
+			};
+		},
+		catch: (e) => J(e)
+	})));
+}), st = (e, t) => m.gen(function* () {
+	let n = h.make(e);
+	if (n?.type !== "object") {
+		let e = n?.anyOf === void 0 ? `type:${JSON.stringify(n?.type ?? null)}` : "anyOf (e.g. Schema.Struct({}) or a union)";
+		return yield* m.fail(new q({
+			reason: "generation",
+			message: `${t}: the output schema must render as a top-level JSON object (type:"object"), but got ${e}. Use a Schema.Struct({ ... }) as the output schema — wrap arrays/scalars in a field (e.g. Schema.Struct({ items: Schema.Array(...) })).`
+		}));
+	}
+	return n;
+}), ct = (e, t) => st(e, t), lt = (e) => at(e.provider ?? n(), e.fallbacks, (n) => G(U(n, "generateObject"), (e) => e.usage, m.gen(function* () {
+	let i = yield* m.try({
+		try: () => $e(e.jsonSchema),
+		catch: () => new q({
+			reason: "decode",
+			message: "generateObjectJson: invalid or unsupported output JSON Schema; no provider call was made"
+		})
+	}), a = rt(n.name, t(n)), s = a.model, c = yield* m.tryPromise({
+		try: () => o({
+			model: s,
+			schema: i,
+			prompt: e.prompt,
+			...e.system === void 0 ? {} : { system: e.system },
+			...e.maxTokens === void 0 ? {} : { maxOutputTokens: e.maxTokens },
+			...e.temperature === void 0 ? {} : { temperature: e.temperature },
+			...e.providerOptions === void 0 ? {} : { providerOptions: e.providerOptions }
+		}),
+		catch: (e) => r.isInstance(e) ? new q({
+			reason: "decode",
+			message: "model output failed JSON Schema validation",
+			evidence: a.evidence(e.usage)
+		}) : J(e, {
+			prefix: "structured provider call failed",
+			evidence: K(n.name, s.modelId, void 0)
+		})
+	});
+	return {
+		object: c.object,
+		evidence: K(n.name, s.modelId, c.usage, c.providerMetadata),
+		usage: {
+			inputTokens: c.usage.inputTokens,
+			outputTokens: c.usage.outputTokens,
+			totalTokens: c.usage.totalTokens
+		},
+		providerMetadata: c.providerMetadata
+	};
+}))), ut = (e) => m.gen(function* () {
+	let t = yield* lt({
+		jsonSchema: yield* st(e.schema, "generateObject"),
+		prompt: e.prompt,
+		...e.system === void 0 ? {} : { system: e.system },
+		...e.provider === void 0 ? {} : { provider: e.provider },
+		...e.fallbacks === void 0 ? {} : { fallbacks: e.fallbacks },
+		...e.maxTokens === void 0 ? {} : { maxTokens: e.maxTokens },
+		...e.temperature === void 0 ? {} : { temperature: e.temperature },
+		...e.providerOptions === void 0 ? {} : { providerOptions: e.providerOptions }
+	}), n = yield* _.decodeUnknown(e.schema)(t.object).pipe(m.mapError(() => new q({
+		reason: "decode",
+		message: "model output failed schema decode",
+		evidence: t.evidence
+	})));
+	return {
+		...t,
+		object: n
+	};
+}), dt = (e) => m.gen(function* () {
+	let a = yield* m.runtime(), o = yield* st(e.schema, "generateObjectWithTools"), u = {
+		store: void 0,
+		...e.toolContext,
+		...e.abortSignal === void 0 ? {} : { abortSignal: e.abortSignal },
+		runtime: a
+	}, d = k(e.tools, u);
+	return yield* at(e.provider ?? n(), e.fallbacks, (n) => G(U(n, "generateObjectWithTools"), (e) => e.usage, m.gen(function* () {
+		let a = rt(n.name, t(n)), u = a.model, f = yield* m.tryPromise({
+			try: () => s({
+				model: u,
+				prompt: e.prompt,
+				tools: d,
+				...e.abortSignal === void 0 ? {} : { abortSignal: e.abortSignal },
+				stopWhen: l(e.maxSteps ?? 8),
+				output: i.object({ schema: c(o) }),
+				...e.system === void 0 ? {} : { system: e.system },
+				...e.maxTokens === void 0 ? {} : { maxOutputTokens: e.maxTokens },
+				...e.temperature === void 0 ? {} : { temperature: e.temperature },
+				...e.providerOptions === void 0 ? {} : { providerOptions: e.providerOptions }
+			}),
+			catch: (e) => r.isInstance(e) ? new q({
+				reason: "decode",
+				message: "model output failed structured decode",
+				evidence: a.evidence(e.usage)
+			}) : J(e, {
+				prefix: "structured provider call failed",
+				evidence: K(n.name, u.modelId, void 0)
+			})
+		}), p = f.totalUsage, h = a.evidence(p), g = yield* m.try({
+			try: () => f.output,
+			catch: () => new q({
+				reason: "decode",
+				message: "model output failed structured decode",
+				evidence: h
+			})
+		});
+		return {
+			object: yield* _.decodeUnknown(e.schema)(g).pipe(m.mapError(() => new q({
+				reason: "decode",
+				message: "model output failed schema decode",
+				evidence: h
+			}))),
+			evidence: h,
+			usage: {
+				inputTokens: p.inputTokens,
+				outputTokens: p.outputTokens,
+				totalTokens: p.totalTokens
+			},
+			providerMetadata: f.providerMetadata
+		};
+	})));
+}), ft = class extends f.Tag("@voltro/AiService")() {}, pt = {
+	generateText: (e) => ot(e),
+	generateObject: (e) => ut(e)
+}, mt = async (e, t) => {
+	if ("bytes" in e) return {
+		audio: e.bytes,
+		mediaType: e.mediaType
+	};
+	if ("url" in e) {
+		let t = await fetch(e.url);
+		if (!t.ok) throw Error(`transcribe: GET ${e.url} answered ${t.status}.`);
+		let n = t.headers.get("content-type") ?? void 0;
+		return {
+			audio: new Uint8Array(await t.arrayBuffer()),
+			...n === void 0 ? {} : { mediaType: n }
+		};
+	}
+	if (t === void 0) throw Error("transcribe: `source.storageRef` needs a `resolveRef` to turn the reference into bytes.\n  → pass `resolveRef: (ref) => storage.read(ref)`. It is injected rather than imported so @voltro/ai does not depend on storage; the app owns tenancy and signing for that read.");
+	let { bytes: n, mediaType: r } = await t(e.storageRef);
+	return {
+		audio: n,
+		mediaType: r
+	};
+}, ht = (e) => {
+	if (!Array.isArray(e) || e.length === 0) return;
+	let t = [];
+	for (let n of e) {
+		let e = typeof n.startSecond == "number" ? n.startSecond : typeof n.start == "number" ? n.start : void 0, r = typeof n.endSecond == "number" ? n.endSecond : typeof n.end == "number" ? n.end : void 0, i = typeof n.text == "string" ? n.text : void 0;
+		if (e === void 0 || r === void 0 || i === void 0) continue;
+		let a = typeof n.speaker == "string" ? n.speaker : void 0;
+		t.push({
+			start: e,
+			end: r,
+			text: i,
+			...a === void 0 ? {} : { speaker: a }
+		});
+	}
+	return t.length > 0 ? t : void 0;
+}, gt = (t) => {
+	let r = t.provider ?? n();
+	return G(U(r, "transcribe"), () => ({}), m.tryPromise({
+		try: async () => {
+			let { audio: n, mediaType: i } = await mt(t.source, t.resolveRef), o = await a({
+				model: e(r),
+				audio: n,
+				...i === void 0 ? {} : { mediaType: i },
+				...t.providerOptions === void 0 ? {} : { providerOptions: t.providerOptions }
+			}), s = t.timestamps === "none" ? void 0 : ht(o.segments);
+			return {
+				text: o.text,
+				...o.language === void 0 ? {} : { language: o.language },
+				...o.durationInSeconds === void 0 ? {} : { durationSec: o.durationInSeconds },
+				...s === void 0 ? {} : { segments: s },
+				...o.providerMetadata === void 0 ? {} : { providerMetadata: o.providerMetadata }
+			};
+		},
+		catch: (e) => J(e)
+	}));
+}, _t = (e) => `sha256:${de("sha256").update(e).digest("hex").slice(0, 16)}`, vt = class extends _.TaggedError()("PromptRenderError", {
+	promptId: _.String,
+	missing: _.Array(_.String),
+	message: _.String
+}) {}, yt = /\{\{\s*([A-Za-z0-9_.]+)\s*\}\}/g, bt = (e) => {
+	let t = /* @__PURE__ */ new Set();
+	for (let n of e.matchAll(yt)) {
+		let e = n[1];
+		e !== void 0 && t.add(e);
+	}
+	return [...t];
+}, xt = (e, t) => _t(`${e.length}:${e}${(t ?? "").length}:${t ?? ""}`), St = (e) => {
+	let t = xt(e.template, e.system), n = bt(e.template);
+	return {
+		...e,
+		digest: t,
+		variables: n,
+		render: (r = {}) => m.suspend(() => {
+			let i = n.filter((e) => r[e] === void 0);
+			if (i.length > 0) return m.fail(new vt({
+				promptId: e.id,
+				missing: i,
+				message: `prompt '${e.id}' needs ${i.map((e) => `{{${e}}}`).join(", ")}, which ${i.length === 1 ? "was" : "were"} not supplied.`
+			}));
+			let a = e.template.replace(yt, (e, t) => String(r[t]));
+			return m.succeed({
+				promptId: e.id,
+				digest: t,
+				...e.label === void 0 ? {} : { label: e.label },
+				prompt: a,
+				...e.system === void 0 ? {} : { system: e.system },
+				definition: e
+			});
+		})
+	};
+}, Ct = (e) => typeof e == "object" && !!e && typeof e.prompt == "string" && typeof e.digest == "string" && typeof e.promptId == "string", Y = "_voltro_prompts", X = E(Y, {
+	id: x({ prefix: "prm" }),
+	promptId: D().maxLength(191),
+	digest: D().maxLength(191),
+	revision: S().default(1),
+	label: D().nullable(),
+	template: D().nullable(),
+	system: D().nullable(),
+	templateChars: S().default(0),
+	firstSeenAt: O().default("now"),
+	lastUsedAt: O().default("now")
+}).unique("byPromptsVersion", ["promptId", "digest"]).index(["promptId"]), wt = 8760, Tt = () => {
+	T({
+		source: "framework",
+		table: Y,
+		timeColumn: "lastUsedAt",
+		ttlMs: oe(process.env.VOLTRO_AI_PROMPTS_TTL_HOURS, wt),
+		envVar: "VOLTRO_AI_PROMPTS_TTL_HOURS",
+		label: "ai prompt versions"
+	});
+}, Z = (e) => ({
+	promptId: String(e.promptId),
+	digest: String(e.digest),
+	revision: Number(e.revision) || 1,
+	label: e.label ?? null,
+	template: e.template ?? null,
+	system: e.system ?? null,
+	firstSeenAt: e.firstSeenAt ?? null,
+	lastUsedAt: e.lastUsedAt ?? null
+}), Et = (e, t) => m.gen(function* () {
+	Tt();
+	let n = t.digest ?? St(t).digest;
+	return yield* m.promise(async () => {
+		let r = (await e.query(w(X).where(v(y("promptId", t.id), y("digest", n))).descriptor))[0];
+		if (r !== void 0) {
+			let t = /* @__PURE__ */ new Date();
+			return await e.updateMany(Y, { lastUsedAt: t }, { where: y("id", String(r.id)) }), Z({
+				...r,
+				lastUsedAt: t
+			});
+		}
+		let i = (await e.query(w(X).where(y("promptId", t.id)).descriptor)).reduce((e, t) => Math.max(e, Number(t.revision) || 0), 0) + 1, a = /* @__PURE__ */ new Date(), o = {
+			id: b({
+				kind: "typeid",
+				prefix: "prm"
+			}, Y),
+			promptId: t.id,
+			digest: n,
+			revision: i,
+			label: t.label ?? null,
+			template: t.template,
+			system: t.system ?? null,
+			templateChars: t.template.length,
+			firstSeenAt: a,
+			lastUsedAt: a
+		};
+		return Z(await e.insertIgnore(Y, o, { conflictColumns: ["promptId", "digest"] }) ?? o);
+	});
+}), Dt = (e, t) => m.promise(async () => {
+	let n = (await e.query(w(X).where(y("digest", t)).descriptor))[0];
+	return n === void 0 ? null : Z(n);
+}), Ot = (e, t) => m.promise(async () => [...await e.query(w(X).where(y("promptId", t)).orderBy("revision", "desc").descriptor)].sort((e, t) => (Number(t.revision) || 0) - (Number(e.revision) || 0)).map(Z)), kt = "_voltro_ai_inferences", At = E(kt, {
+	id: x({ prefix: "aiinf" }),
+	tenantId: D().nullable(),
+	executionId: D(),
+	workflowName: D(),
+	runId: D().nullable(),
+	signalName: D(),
+	stepName: D(),
+	status: D().oneOf([
+		"pending",
+		"running",
+		"succeeded",
+		"failed"
+	]).default("pending"),
+	kind: D().oneOf([
+		"text",
+		"object",
+		"transcription",
+		"media"
+	]).default("text"),
+	provider: D(),
+	model: D(),
+	prompt: D(),
+	system: D().nullable(),
+	promptId: D().nullable(),
+	promptDigest: D().nullable(),
+	promptRevision: S().nullable(),
+	maxTokens: S().nullable(),
+	providerOptions: C().nullable(),
+	jsonSchema: C().nullable(),
+	transcription: C().nullable(),
+	media: C().nullable(),
+	taskHandle: C().nullable(),
+	taskSubmittedAt: O().nullable(),
+	result: C().nullable(),
+	errorMessage: D().nullable(),
+	attempts: S().default(0),
+	fence: S().default(0),
+	leaseUntil: O().nullable(),
+	claimedBy: D().nullable(),
+	enqueuedAt: O().default("now"),
+	startedAt: O().nullable(),
+	completedAt: O().nullable(),
+	durationMs: S().nullable()
+}).index(["status", "enqueuedAt"]).index(["status", "leaseUntil"]).index(["executionId"]).index(["tenantId"]), jt = 3e5, Mt = 4, Q = (e, t) => {
+	let n = typeof e == "number" ? e : Number(e);
+	return Number.isFinite(n) ? Math.trunc(n) : t;
+}, $ = (e) => e == null ? null : String(e), Nt = (e) => {
+	if (e instanceof Date) return e;
+	if (typeof e == "string" || typeof e == "number") {
+		let t = new Date(e);
+		return Number.isNaN(t.getTime()) ? null : t;
+	}
+	return null;
+}, Pt = (e) => ({
+	id: String(e.id),
+	tenantId: $(e.tenantId),
+	executionId: String(e.executionId),
+	workflowName: String(e.workflowName),
+	runId: $(e.runId),
+	signalName: String(e.signalName),
+	stepName: String(e.stepName),
+	kind: e.kind === "object" ? "object" : e.kind === "transcription" ? "transcription" : e.kind === "media" ? "media" : "text",
+	provider: String(e.provider),
+	model: String(e.model),
+	prompt: String(e.prompt ?? ""),
+	system: $(e.system),
+	promptId: $(e.promptId),
+	promptDigest: $(e.promptDigest),
+	promptRevision: e.promptRevision === null || e.promptRevision === void 0 ? null : Q(e.promptRevision, 0),
+	maxTokens: e.maxTokens === null || e.maxTokens === void 0 ? null : Q(e.maxTokens, 0),
+	providerOptions: e.providerOptions ?? null,
+	jsonSchema: e.jsonSchema ?? null,
+	transcription: e.transcription ?? null,
+	media: e.media ?? null,
+	taskHandle: typeof e.taskHandle == "object" ? e.taskHandle : null,
+	taskSubmittedAt: Nt(e.taskSubmittedAt),
+	attempts: Q(e.attempts, 0),
+	fence: Q(e.fence, 0)
+}), Ft = (e) => {
+	let t = e.attempts + 1;
+	return t >= e.maxAttempts ? {
+		kind: "give-up",
+		attempts: t
+	} : {
+		kind: "retry",
+		attempts: t,
+		delayMs: e.retryAfterMs !== void 0 && e.retryAfterMs > 0 ? e.retryAfterMs : Math.min(e.baseDelayMs * 2 ** (t - 1), e.maxDelayMs)
+	};
+}, It = (e) => `voltro.ai.infer:${e}`, Lt = (e, t) => `aiinf_${de("sha256").update(`${e}\u0000${t}`).digest("hex").slice(0, 40)}`;
+//#endregion
+export { M as $, tt as A, Ie as B, J as C, dt as D, lt as E, Be as F, Pe as G, z as H, ze as I, Me as J, He as K, je as L, Qe as M, G as N, ot as O, B as P, j as Q, Ge as R, ft as S, ut as T, Ne as U, Re as V, Fe as W, N as X, P as Y, A as Z, Ot as _, Ft as a, me as at, gt as b, Pt as c, vt as d, Se as et, St as f, xt as g, Dt as h, At as i, k as it, U as j, ct as k, wt as l, _t as m, Mt as n, Ae as nt, It as o, he as ot, Ct as p, We as q, jt as r, ke as rt, Lt as s, ge as st, kt as t, F as tt, Y as u, X as v, pt as w, q as x, Et as y, Le as z };