@voltro/ai 0.57.0 → 0.58.0

package/dist/inferenceQueue-EQJJepZ8.js

@@ -0,0 +1,632 @@
+import { a as e, r as t, s as n } from "./provider-BOgeCRkg.js";
+import { Output as r, experimental_transcribe as i, generateObject as a, generateText as o, jsonSchema as s, stepCountIs as c, tool as l } from "ai";
+import { Context as u, Effect as d, JSONSchema as ee, Metric as f, MetricBoundaries as te, Runtime as p, Schema as m, Stream as ne } from "effect";
+import { and as h, eq as g, generateId as re, gte as ie, id as _, integer as v, json as y, queryFor as b, registerRetention as x, retentionTtlMsFromEnv as S, table as C, text as w, timestamp as T } from "@voltro/database";
+import { createHash as E } from "node:crypto";
+//#region src/tools.ts
+var ae = (e, t) => e(t), D = Symbol.for("@voltro/ai/Tool"), oe = (e) => ({
+	...e,
+	[D]: !0
+}), O = (e) => typeof e == "object" && !!e && e[D] === !0, se = (e) => {
+	let t = e.default;
+	if (typeof t != "function") return 0;
+	let n = 0;
+	for (let r of Object.values(e)) O(r) && r.execute === void 0 && r.handler === void 0 && (r.handler = t, n += 1);
+	return n;
+}, k = (e, t) => {
+	let n = t ?? p.defaultRuntime;
+	return p.runPromise(n)(e);
+}, A = (e, t, n) => n ? e.pipe(d.timeoutFail({
+	duration: `${n} millis`,
+	onTimeout: () => /* @__PURE__ */ Error(`tool '${t}' timed out`)
+})) : e, ce = (e, t, n) => {
+	if (e.execute) return k(A(ae(e.execute, t), e.name, e.timeoutMs), n.runtime);
+	if (e.handler) {
+		let r = e.handler, i = r(t, n);
+		return d.isEffect(i) ? k(A(i, e.name, e.timeoutMs), n.runtime) : i;
+	}
+	return Promise.reject(/* @__PURE__ */ Error(`tool '${e.name}' has neither execute nor handler`));
+}, le = (e) => {
+	let t = ee.make(e);
+	return t && t.type === "object" ? t : {
+		type: "object",
+		properties: {},
+		additionalProperties: !1
+	};
+}, ue = (e) => {
+	let t = e.inputJsonSchema;
+	return t === void 0 ? le(e.input) : t.type === "object" ? t : {
+		type: "object",
+		properties: {},
+		additionalProperties: !0
+	};
+}, j = (e, t = { store: void 0 }) => {
+	let n = {};
+	for (let [r, i] of Object.entries(e)) n[i.name || r] = l({
+		description: i.description,
+		inputSchema: s(ue(i)),
+		execute: async (e) => {
+			let n = await d.runPromise(m.decodeUnknown(i.input)(e)), r = await ce(i, n, t);
+			return i.output ? d.runPromise(m.decodeUnknown(i.output)(r)) : r;
+		}
+	});
+	return n;
+}, de = {
+	"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
+	}
+}, M = {}, fe = (e) => {
+	let t = M;
+	return M = { ...e }, t;
+}, pe = () => ({
+	...de,
+	...M
+}), me = (e) => {
+	let t = pe();
+	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];
+	}
+}, N = (e, t) => {
+	let n = e.inputTokens ?? 0, r = e.outputTokens ?? 0;
+	if (t.actualCostUsd !== void 0) {
+		let e = n + r, i = e > 0 ? n / e : 0;
+		return {
+			inputTokens: n,
+			outputTokens: r,
+			inputCostUsd: t.actualCostUsd * i,
+			outputCostUsd: t.actualCostUsd * (1 - i),
+			totalCostUsd: t.actualCostUsd,
+			costSource: "gateway"
+		};
+	}
+	let i = t.price ?? me(t.model) ?? {
+		inputPer1M: 0,
+		outputPer1M: 0
+	}, a = n / 1e6 * i.inputPer1M, o = r / 1e6 * i.outputPer1M;
+	return {
+		inputTokens: n,
+		outputTokens: r,
+		inputCostUsd: a,
+		outputCostUsd: o,
+		totalCostUsd: a + o,
+		costSource: "estimated"
+	};
+}, P = "_voltro_ai_usage", he = C(P, {
+	id: _({ prefix: "aiu" }),
+	tenantId: w().nullable(),
+	provider: w(),
+	model: w(),
+	operation: w(),
+	agent: w().nullable(),
+	inputTokens: v().default(0),
+	outputTokens: v().default(0),
+	costMicroUsd: v().default(0),
+	costSource: w().default("estimated"),
+	promptId: w().nullable(),
+	promptDigest: w().nullable(),
+	promptRevision: v().nullable(),
+	calledAt: T().default("now")
+}).index(["tenantId"]).index(["promptDigest"]), ge = (e) => {
+	let t = e?.gateway?.cost, n = typeof t == "number" ? t : typeof t == "string" ? Number(t) : NaN;
+	return Number.isFinite(n) ? n : void 0;
+}, _e = (e, t) => (x({
+	source: "plugin",
+	table: P,
+	timeColumn: "calledAt",
+	ttlMs: S(process.env.VOLTRO_AI_USAGE_TTL_HOURS, 8760)
+}), d.promise(async () => {
+	let n = N(t.usage, {
+		model: t.model,
+		...t.price === void 0 ? {} : { price: t.price },
+		...t.actualCostUsd === void 0 ? {} : { actualCostUsd: t.actualCostUsd }
+	});
+	return await e.insert(P, {
+		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,
+		promptId: t.prompt?.promptId ?? null,
+		promptDigest: t.prompt?.digest ?? null,
+		promptRevision: t.prompt?.revision ?? null,
+		calledAt: t.calledAt ?? /* @__PURE__ */ new Date()
+	}), n;
+})), ve = (e, t = {}) => d.promise(async () => {
+	let n = [
+		...t.tenantId === void 0 ? [] : [g("tenantId", t.tenantId)],
+		...t.since === void 0 ? [] : [ie("calledAt", t.since)],
+		...t.promptDigest === void 0 ? [] : [g("promptDigest", t.promptDigest)]
+	], r = b(he), i = n.length === 0 ? r : r.where(n.length === 1 ? n[0] : h(...n));
+	return (await e.query(i.descriptor)).reduce((e, t) => e + (Number(t.costMicroUsd) || 0), 0) / 1e6;
+}), ye = class extends m.TaggedError()("AiBudgetExceeded", {
+	tenantId: m.NullOr(m.String),
+	limitUsd: m.Number,
+	spentUsd: m.Number,
+	attemptedUsd: m.Number
+}) {}, F = "_voltro_ai_budget", I = C(F, {
+	id: _({ prefix: "aib" }),
+	bucket: w().maxLength(191),
+	tenantId: w().nullable(),
+	reservedMicroUsd: v().default(0),
+	updatedAt: T().default("now")
+}).unique("byAiBudgetBucket", ["bucket"]), be = (e, t) => `${e ?? "global"}::${t === void 0 ? "all" : String(t.getTime())}`, L = 32, xe = (e, t) => d.gen(function* () {
+	x({
+		source: "plugin",
+		table: F,
+		timeColumn: "updatedAt",
+		ttlMs: S(process.env.VOLTRO_AI_USAGE_TTL_HOURS, 8760)
+	});
+	let n = t.addUsd ?? 0, r = Math.round(n * 1e6), i = Math.round(t.limitUsd * 1e6), a = be(t.tenantId, t.since), o = t.tenantId ?? null, s = (e) => new ye({
+		tenantId: o,
+		limitUsd: t.limitUsd,
+		spentUsd: e / 1e6,
+		attemptedUsd: n
+	}), c = yield* d.promise(async () => {
+		for (let t = 0; t < L; t++) {
+			t > 0 && await new Promise((e) => setTimeout(e, t));
+			let n = (await e.query(b(I).where(g("bucket", a)).descriptor))[0], s = n && Number(n.reservedMicroUsd) || 0;
+			if (s + r > i) return {
+				ok: !1,
+				reservedMicro: s
+			};
+			if (r === 0) return { ok: !0 };
+			if (n) {
+				if (await e.updateMany("_voltro_ai_budget", {
+					reservedMicroUsd: s + r,
+					updatedAt: /* @__PURE__ */ new Date()
+				}, { where: h(g("id", String(n.id)), g("reservedMicroUsd", s)) }) === 1) return { ok: !0 };
+			} else {
+				let t = re({
+					kind: "typeid",
+					prefix: "aib"
+				}, F), n = await e.insertIgnore(F, {
+					id: t,
+					bucket: a,
+					tenantId: o,
+					reservedMicroUsd: r,
+					updatedAt: /* @__PURE__ */ new Date()
+				}, { conflictColumns: ["bucket"] });
+				if (String(n.id) === t) return { ok: !0 };
+			}
+		}
+		throw Error(`requireAiBudget: contention on bucket '${a}' exceeded ${L} attempts`);
+	});
+	if (!c.ok) return yield* d.fail(s(c.reservedMicro));
+}), Se = te.exponential({
+	start: .001,
+	factor: 2,
+	count: 16
+}), R = f.counter("voltro_ai_calls_total", { description: "Total AI provider calls (generate + stream)." }), z = f.counter("voltro_ai_call_errors_total", { description: "AI provider calls that errored." }), B = f.histogram("voltro_ai_call_duration_seconds", Se, "AI provider call duration in seconds."), Ce = f.counter("voltro_ai_input_tokens_total", {
+	description: "Total prompt (input) tokens across AI calls.",
+	incremental: !0
+}), we = f.counter("voltro_ai_output_tokens_total", {
+	description: "Total completion (output) tokens across AI calls.",
+	incremental: !0
+}), Te = f.counter("voltro_ai_cost_microusd_total", {
+	description: "Total estimated AI spend in micro-USD (USD × 1e6).",
+	incremental: !0
+}), V = (e, t) => ({
+	provider: e.name,
+	model: e.model ?? e.name,
+	operation: t
+}), H = (e, t, n) => e.pipe(d.tagMetrics("provider", t.provider), d.tagMetrics("model", t.model), d.tagMetrics("operation", t.operation), d.tagMetrics("status", n)), U = (e, t) => {
+	let n = t.inputTokens ?? 0, r = t.outputTokens ?? 0, i = N({
+		inputTokens: n,
+		outputTokens: r
+	}, { model: e.model }), a = Math.round(i.totalCostUsd * 1e6);
+	return d.all([
+		H(f.incrementBy(Ce, n), e, "ok"),
+		H(f.incrementBy(we, r), e, "ok"),
+		H(f.incrementBy(Te, a), e, "ok")
+	], { discard: !0 });
+}, W = (e, t, n) => d.gen(function* () {
+	let r = yield* d.clockWith((e) => e.currentTimeMillis);
+	return yield* n.pipe(d.tap((n) => U(e, t(n))), d.tap(() => H(f.increment(R), e, "ok").pipe(d.zipRight(d.clockWith((e) => e.currentTimeMillis).pipe(d.flatMap((t) => H(f.update(B, (t - r) / 1e3), e, "ok")))))), d.tapError(() => H(f.increment(R), e, "error").pipe(d.zipRight(H(f.increment(z), e, "error")), d.zipRight(d.clockWith((e) => e.currentTimeMillis).pipe(d.flatMap((t) => H(f.update(B, (t - r) / 1e3), e, "error")))))), d.withSpan("voltro.ai.call", { attributes: {
+		"ai.provider": e.provider,
+		"ai.model": e.model,
+		"ai.operation": e.operation
+	} }));
+}), Ee = (e, t) => {
+	let n = Date.now();
+	return t.pipe(ne.tap((t) => t._tag === "done" ? H(f.increment(R), e, "ok").pipe(d.zipRight(H(f.update(B, (Date.now() - n) / 1e3), e, "ok")), d.zipRight(U(e, t.usage ?? {}))) : t._tag === "error" ? H(f.increment(R), e, "error").pipe(d.zipRight(H(f.increment(z), e, "error")), d.zipRight(H(f.update(B, (Date.now() - n) / 1e3), e, "error"))) : d.void), ne.withSpan("voltro.ai.call", { attributes: {
+		"ai.provider": e.provider,
+		"ai.model": e.model,
+		"ai.operation": e.operation
+	} }));
+}, G = class extends m.TaggedError()("AiError", {
+	reason: m.Literal("generation", "decode"),
+	message: m.String
+}) {}, K = (e, t, n) => {
+	let r = [e, ...t ?? []], i = (e) => n(r[e]).pipe(d.catchTag("AiError", (t) => t.reason === "generation" && e + 1 < r.length ? i(e + 1) : d.fail(t)));
+	return i(0);
+}, De = (n) => K(n.provider ?? t(), n.fallbacks, (t) => W(V(t, "generateText"), (e) => e.usage, d.tryPromise({
+	try: async () => {
+		let r = e(t), i = await o({
+			model: r,
+			prompt: n.prompt,
+			...n.system === void 0 ? {} : { system: n.system },
+			...n.maxTokens === void 0 ? {} : { maxOutputTokens: n.maxTokens },
+			...n.providerOptions === void 0 ? {} : { providerOptions: n.providerOptions }
+		});
+		return {
+			text: i.text,
+			usage: {
+				inputTokens: i.usage.inputTokens,
+				outputTokens: i.usage.outputTokens,
+				totalTokens: i.usage.totalTokens
+			},
+			providerMetadata: i.providerMetadata
+		};
+	},
+	catch: (e) => new G({
+		reason: "generation",
+		message: String(e?.message ?? e)
+	})
+}))), q = (e, t) => d.gen(function* () {
+	let n = ee.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* d.fail(new G({
+			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;
+}), Oe = (e, t) => q(e, t), ke = (n) => K(n.provider ?? t(), n.fallbacks, (t) => W(V(t, "generateObject"), (e) => e.usage, d.gen(function* () {
+	let r = e(t), i = yield* d.tryPromise({
+		try: () => a({
+			model: r,
+			schema: s(n.jsonSchema),
+			prompt: n.prompt,
+			...n.system === void 0 ? {} : { system: n.system },
+			...n.maxTokens === void 0 ? {} : { maxOutputTokens: n.maxTokens },
+			...n.providerOptions === void 0 ? {} : { providerOptions: n.providerOptions }
+		}),
+		catch: (e) => new G({
+			reason: "generation",
+			message: String(e?.message ?? e)
+		})
+	});
+	return {
+		object: i.object,
+		usage: {
+			inputTokens: i.usage.inputTokens,
+			outputTokens: i.usage.outputTokens,
+			totalTokens: i.usage.totalTokens
+		},
+		providerMetadata: i.providerMetadata
+	};
+}))), Ae = (e) => d.gen(function* () {
+	let t = yield* ke({
+		jsonSchema: yield* q(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.providerOptions === void 0 ? {} : { providerOptions: e.providerOptions }
+	}), n = yield* m.decodeUnknown(e.schema)(t.object).pipe(d.mapError((e) => new G({
+		reason: "decode",
+		message: `model output failed schema decode: ${String(e)}`
+	})));
+	return {
+		...t,
+		object: n
+	};
+}), je = (n) => d.gen(function* () {
+	let i = yield* d.runtime(), a = yield* q(n.schema, "generateObjectWithTools"), l = {
+		store: void 0,
+		...n.toolContext,
+		runtime: i
+	}, u = j(n.tools, l);
+	return yield* K(n.provider ?? t(), n.fallbacks, (t) => W(V(t, "generateObjectWithTools"), (e) => e.usage, d.gen(function* () {
+		let i = e(t), l = yield* d.tryPromise({
+			try: () => o({
+				model: i,
+				prompt: n.prompt,
+				tools: u,
+				stopWhen: c(n.maxSteps ?? 8),
+				output: r.object({ schema: s(a) }),
+				...n.system === void 0 ? {} : { system: n.system },
+				...n.maxTokens === void 0 ? {} : { maxOutputTokens: n.maxTokens },
+				...n.providerOptions === void 0 ? {} : { providerOptions: n.providerOptions }
+			}),
+			catch: (e) => new G({
+				reason: "generation",
+				message: String(e?.message ?? e)
+			})
+		});
+		return {
+			object: yield* m.decodeUnknown(n.schema)(l.output).pipe(d.mapError((e) => new G({
+				reason: "decode",
+				message: `model output failed schema decode: ${String(e)}`
+			}))),
+			usage: {
+				inputTokens: l.usage.inputTokens,
+				outputTokens: l.usage.outputTokens,
+				totalTokens: l.usage.totalTokens
+			},
+			providerMetadata: l.providerMetadata
+		};
+	})));
+}), Me = class extends u.Tag("@voltro/AiService")() {}, Ne = {
+	generateText: (e) => De(e),
+	generateObject: (e) => Ae(e)
+}, Pe = 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
+	};
+}, Fe = (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;
+}, Ie = (e) => {
+	let r = e.provider ?? t();
+	return W(V(r, "transcribe"), () => ({}), d.tryPromise({
+		try: async () => {
+			let { audio: t, mediaType: a } = await Pe(e.source, e.resolveRef), o = await i({
+				model: n(r),
+				audio: t,
+				...a === void 0 ? {} : { mediaType: a },
+				...e.providerOptions === void 0 ? {} : { providerOptions: e.providerOptions }
+			}), s = e.timestamps === "none" ? void 0 : Fe(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) => new G({
+			reason: "generation",
+			message: String(e?.message ?? e)
+		})
+	}));
+}, Le = (e) => `sha256:${E("sha256").update(e).digest("hex").slice(0, 16)}`, Re = class extends m.TaggedError()("PromptRenderError", {
+	promptId: m.String,
+	missing: m.Array(m.String),
+	message: m.String
+}) {}, J = /\{\{\s*([A-Za-z0-9_.]+)\s*\}\}/g, ze = (e) => {
+	let t = /* @__PURE__ */ new Set();
+	for (let n of e.matchAll(J)) {
+		let e = n[1];
+		e !== void 0 && t.add(e);
+	}
+	return [...t];
+}, Be = (e, t) => Le(`${e.length}:${e}${(t ?? "").length}:${t ?? ""}`), Ve = (e) => {
+	let t = Be(e.template, e.system), n = ze(e.template);
+	return {
+		...e,
+		digest: t,
+		variables: n,
+		render: (r = {}) => d.suspend(() => {
+			let i = n.filter((e) => r[e] === void 0);
+			if (i.length > 0) return d.fail(new Re({
+				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(J, (e, t) => String(r[t]));
+			return d.succeed({
+				promptId: e.id,
+				digest: t,
+				...e.label === void 0 ? {} : { label: e.label },
+				prompt: a,
+				...e.system === void 0 ? {} : { system: e.system },
+				definition: e
+			});
+		})
+	};
+}, He = (e) => typeof e == "object" && !!e && typeof e.prompt == "string" && typeof e.digest == "string" && typeof e.promptId == "string", Y = "_voltro_prompts", X = C(Y, {
+	id: _({ prefix: "prm" }),
+	promptId: w().maxLength(191),
+	digest: w().maxLength(191),
+	revision: v().default(1),
+	label: w().nullable(),
+	template: w().nullable(),
+	system: w().nullable(),
+	templateChars: v().default(0),
+	firstSeenAt: T().default("now"),
+	lastUsedAt: T().default("now")
+}).unique("byPromptsVersion", ["promptId", "digest"]).index(["promptId"]), Ue = 8760, We = () => {
+	x({
+		source: "framework",
+		table: Y,
+		timeColumn: "lastUsedAt",
+		ttlMs: S(process.env.VOLTRO_AI_PROMPTS_TTL_HOURS, Ue),
+		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
+}), Ge = (e, t) => d.gen(function* () {
+	We();
+	let n = t.digest ?? Ve(t).digest;
+	return yield* d.promise(async () => {
+		let r = (await e.query(b(X).where(h(g("promptId", t.id), g("digest", n))).descriptor))[0];
+		if (r !== void 0) {
+			let t = /* @__PURE__ */ new Date();
+			return await e.updateMany(Y, { lastUsedAt: t }, { where: g("id", String(r.id)) }), Z({
+				...r,
+				lastUsedAt: t
+			});
+		}
+		let i = (await e.query(b(X).where(g("promptId", t.id)).descriptor)).reduce((e, t) => Math.max(e, Number(t.revision) || 0), 0) + 1, a = /* @__PURE__ */ new Date(), o = {
+			id: re({
+				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);
+	});
+}), Ke = (e, t) => d.promise(async () => {
+	let n = (await e.query(b(X).where(g("digest", t)).descriptor))[0];
+	return n === void 0 ? null : Z(n);
+}), qe = (e, t) => d.promise(async () => [...await e.query(b(X).where(g("promptId", t)).orderBy("revision", "desc").descriptor)].sort((e, t) => (Number(t.revision) || 0) - (Number(e.revision) || 0)).map(Z)), Je = "_voltro_ai_inferences", Ye = C(Je, {
+	id: _({ prefix: "aiinf" }),
+	tenantId: w().nullable(),
+	executionId: w(),
+	workflowName: w(),
+	runId: w().nullable(),
+	signalName: w(),
+	stepName: w(),
+	status: w().oneOf([
+		"pending",
+		"running",
+		"succeeded",
+		"failed"
+	]).default("pending"),
+	kind: w().oneOf([
+		"text",
+		"object",
+		"transcription"
+	]).default("text"),
+	provider: w(),
+	model: w(),
+	prompt: w(),
+	system: w().nullable(),
+	promptId: w().nullable(),
+	promptDigest: w().nullable(),
+	promptRevision: v().nullable(),
+	maxTokens: v().nullable(),
+	providerOptions: y().nullable(),
+	jsonSchema: y().nullable(),
+	transcription: y().nullable(),
+	result: y().nullable(),
+	errorMessage: w().nullable(),
+	attempts: v().default(0),
+	leaseUntil: T().nullable(),
+	claimedBy: w().nullable(),
+	enqueuedAt: T().default("now"),
+	startedAt: T().nullable(),
+	completedAt: T().nullable(),
+	durationMs: v().nullable()
+}).index(["status", "enqueuedAt"]).index(["status", "leaseUntil"]).index(["executionId"]).index(["tenantId"]), Xe = 3e5, Ze = 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), Qe = (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" : "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,
+	attempts: Q(e.attempts, 0)
+}), $e = (e) => {
+	let t = e.attempts + 1;
+	return t >= e.maxAttempts ? {
+		kind: "give-up",
+		attempts: t
+	} : {
+		kind: "retry",
+		attempts: t,
+		delayMs: e.retryAfterMs !== void 0 && e.retryAfterMs > 0 ? e.retryAfterMs : Math.min(e.baseDelayMs * 2 ** (t - 1), e.maxDelayMs)
+	};
+}, et = (e) => `voltro.ai.infer:${e}`, tt = (e, t) => `aiinf_${E("sha256").update(`${e}\u0000${t}`).digest("hex").slice(0, 40)}`;
+//#endregion
+export { Ee as A, ge as B, Ne as C, De as D, je as E, de as F, fe as G, me as H, I, O as J, j as K, ve as L, F as M, P as N, Oe as O, ye as P, he as R, Me as S, ke as T, _e as U, pe as V, xe as W, se as Y, qe as _, $e as a, Ie as b, Qe as c, Re as d, Ve as f, Be as g, Ke as h, Ye as i, W as j, V as k, Ue as l, Le as m, Ze as n, et as o, He as p, oe as q, Xe as r, tt as s, Je as t, Y as u, X as v, Ae as w, G as x, Ge as y, N as z };