@voltro/ai 0.85.0 → 0.87.0

package/dist/observe.js

@@ -0,0 +1,52 @@
+import { estimateCostUsd as e } from "./cost.js";
+import { Effect as t, Metric as n, Stream as r } from "effect";
+//#region src/observe.ts
+var i = Array.from({ length: 16 }, (e, t) => .001 * 2 ** t), a = n.counter("voltro_ai_calls_total", { description: "Total AI provider calls (generate + stream)." }), o = n.counter("voltro_ai_call_errors_total", { description: "AI provider calls that errored." }), s = n.histogram("voltro_ai_call_duration_seconds", {
+	boundaries: i,
+	description: "AI provider call duration in seconds."
+}), c = n.counter("voltro_ai_input_tokens_total", {
+	description: "Total prompt (input) tokens across AI calls.",
+	incremental: !0
+}), l = n.counter("voltro_ai_output_tokens_total", {
+	description: "Total completion (output) tokens across AI calls.",
+	incremental: !0
+}), u = n.counter("voltro_ai_cost_microusd_total", {
+	description: "Total estimated AI spend in micro-USD (USD × 1e6).",
+	incremental: !0
+}), d = (e, t) => ({
+	provider: e.name,
+	model: e.model ?? e.name,
+	operation: t
+}), f = (e, r, i) => t.updateService(e, n.CurrentMetricAttributes, (e) => ({
+	...e,
+	provider: r.provider,
+	model: r.model,
+	operation: r.operation,
+	status: i
+})), p = (r, i) => {
+	let a = i.inputTokens ?? 0, o = i.outputTokens ?? 0, s = e({
+		inputTokens: a,
+		outputTokens: o
+	}, { model: r.model }), d = Math.round(s.totalCostUsd * 1e6);
+	return t.all([
+		f(n.update(c, a), r, "ok"),
+		f(n.update(l, o), r, "ok"),
+		f(n.update(u, d), r, "ok")
+	], { discard: !0 });
+}, m = (e, r, i) => t.gen(function* () {
+	let c = yield* t.clockWith((e) => e.currentTimeMillis);
+	return yield* i.pipe(t.tap((t) => p(e, r(t))), t.tap(() => f(n.update(a, 1), e, "ok").pipe(t.andThen(t.clockWith((e) => e.currentTimeMillis).pipe(t.flatMap((t) => f(n.update(s, (t - c) / 1e3), e, "ok")))))), t.tapError(() => f(n.update(a, 1), e, "error").pipe(t.andThen(f(n.update(o, 1), e, "error")), t.andThen(t.clockWith((e) => e.currentTimeMillis).pipe(t.flatMap((t) => f(n.update(s, (t - c) / 1e3), e, "error")))))), t.withSpan("voltro.ai.call", { attributes: {
+		"ai.provider": e.provider,
+		"ai.model": e.model,
+		"ai.operation": e.operation
+	} }));
+}), h = (e, i) => {
+	let c = Date.now();
+	return i.pipe(r.tap((r) => r._tag === "done" ? f(n.update(a, 1), e, "ok").pipe(t.andThen(f(n.update(s, (Date.now() - c) / 1e3), e, "ok")), t.andThen(p(e, r.usage ?? {}))) : r._tag === "error" ? f(n.update(a, 1), e, "error").pipe(t.andThen(f(n.update(o, 1), e, "error")), t.andThen(f(n.update(s, (Date.now() - c) / 1e3), e, "error"))) : t.void), r.withSpan("voltro.ai.call", { attributes: {
+		"ai.provider": e.provider,
+		"ai.model": e.model,
+		"ai.operation": e.operation
+	} }));
+};
+//#endregion
+export { d as aiCallAttributes, h as observeStream, m as withAiCallSpan };

package/dist/prompts.js

@@ -0,0 +1,103 @@
+import { Effect as e, Schema as t } from "effect";
+import { and as n, eq as r, generateId as i, id as a, integer as o, queryFor as s, registerRetention as c, retentionTtlMsFromEnv as l, table as u, text as d, timestamp as f } from "@voltro/database";
+import { createHash as p } from "node:crypto";
+//#region src/prompts.ts
+var m = (e) => `sha256:${p("sha256").update(e).digest("hex").slice(0, 16)}`, h = class extends t.TaggedError()("PromptRenderError", {
+	promptId: t.String,
+	missing: t.Array(t.String),
+	message: t.String
+}) {}, g = /\{\{\s*([A-Za-z0-9_.]+)\s*\}\}/g, _ = (e) => {
+	let t = /* @__PURE__ */ new Set();
+	for (let n of e.matchAll(g)) {
+		let e = n[1];
+		e !== void 0 && t.add(e);
+	}
+	return [...t];
+}, v = (e, t) => m(`${e.length}:${e}${(t ?? "").length}:${t ?? ""}`), y = (t) => {
+	let n = v(t.template, t.system), r = _(t.template);
+	return {
+		...t,
+		digest: n,
+		variables: r,
+		render: (i = {}) => e.suspend(() => {
+			let a = r.filter((e) => i[e] === void 0);
+			if (a.length > 0) return e.fail(new h({
+				promptId: t.id,
+				missing: a,
+				message: `prompt '${t.id}' needs ${a.map((e) => `{{${e}}}`).join(", ")}, which ${a.length === 1 ? "was" : "were"} not supplied.`
+			}));
+			let o = t.template.replace(g, (e, t) => String(i[t]));
+			return e.succeed({
+				promptId: t.id,
+				digest: n,
+				...t.label === void 0 ? {} : { label: t.label },
+				prompt: o,
+				...t.system === void 0 ? {} : { system: t.system },
+				definition: t
+			});
+		})
+	};
+}, b = (e) => typeof e == "object" && !!e && typeof e.prompt == "string" && typeof e.digest == "string" && typeof e.promptId == "string", x = "_voltro_prompts", S = u(x, {
+	id: a({ prefix: "prm" }),
+	promptId: d().maxLength(191),
+	digest: d().maxLength(191),
+	revision: o().default(1),
+	label: d().nullable(),
+	template: d().nullable(),
+	system: d().nullable(),
+	templateChars: o().default(0),
+	firstSeenAt: f().default("now"),
+	lastUsedAt: f().default("now")
+}).unique("byPromptsVersion", ["promptId", "digest"]).index(["promptId"]), C = 8760, w = () => {
+	c({
+		source: "framework",
+		table: x,
+		timeColumn: "lastUsedAt",
+		ttlMs: l(process.env.VOLTRO_AI_PROMPTS_TTL_HOURS, C),
+		envVar: "VOLTRO_AI_PROMPTS_TTL_HOURS",
+		label: "ai prompt versions"
+	});
+}, T = (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
+}), E = (t, a) => e.gen(function* () {
+	w();
+	let o = a.digest ?? y(a).digest;
+	return yield* e.promise(async () => {
+		let e = (await t.query(s(S).where(n(r("promptId", a.id), r("digest", o))).descriptor))[0];
+		if (e !== void 0) {
+			let n = /* @__PURE__ */ new Date();
+			return await t.updateMany(x, { lastUsedAt: n }, { where: r("id", String(e.id)) }), T({
+				...e,
+				lastUsedAt: n
+			});
+		}
+		let c = (await t.query(s(S).where(r("promptId", a.id)).descriptor)).reduce((e, t) => Math.max(e, Number(t.revision) || 0), 0) + 1, l = /* @__PURE__ */ new Date(), u = {
+			id: i({
+				kind: "typeid",
+				prefix: "prm"
+			}, x),
+			promptId: a.id,
+			digest: o,
+			revision: c,
+			label: a.label ?? null,
+			template: a.template,
+			system: a.system ?? null,
+			templateChars: a.template.length,
+			firstSeenAt: l,
+			lastUsedAt: l
+		};
+		return T(await t.insertIgnore(x, u, { conflictColumns: ["promptId", "digest"] }) ?? u);
+	});
+}), D = (t, n) => e.promise(async () => {
+	let e = (await t.query(s(S).where(r("digest", n)).descriptor))[0];
+	return e === void 0 ? null : T(e);
+}), O = (t, n) => e.promise(async () => [...await t.query(s(S).where(r("promptId", n)).orderBy("revision", "desc").descriptor)].sort((e, t) => (Number(t.revision) || 0) - (Number(e.revision) || 0)).map(T));
+//#endregion
+export { C as DEFAULT_PROMPT_RETENTION_HOURS, x as PROMPTS_TABLE, h as PromptRenderError, y as definePrompt, b as isRenderedPrompt, m as promptDigest, D as promptVersionByDigest, v as promptVersionDigest, O as promptVersionsFor, S as promptsTable, E as recordPromptVersion };

package/dist/provider.js

@@ -0,0 +1,319 @@
+import { aiSetting as e, isProductionEnv as t } from "./aiSettings.js";
+import { applyMiddleware as n } from "./middleware.js";
+import { boundedGatewaySettings as r, gatewayImageModel as i, gatewayVideoModel as a, mediaResponseBudget as o } from "./gatewayMedia.js";
+import { anthropic as s, createAnthropic as c } from "@ai-sdk/anthropic";
+import { createOpenAI as l, openai as u } from "@ai-sdk/openai";
+import { createGateway as d, gateway as f } from "@ai-sdk/gateway";
+//#region src/provider.ts
+var p = {
+	inputTokens: {
+		total: 0,
+		noCache: 0,
+		cacheRead: 0,
+		cacheWrite: 0
+	},
+	outputTokens: {
+		total: 0,
+		text: 0,
+		reasoning: 0
+	}
+}, m = (e) => ({
+	unified: e,
+	raw: void 0
+}), h = class {
+	turns;
+	cursor = 0;
+	calls = [];
+	constructor(e) {
+		this.turns = e.length > 0 ? e : [{ text: "" }];
+	}
+	record(e, t) {
+		this.calls.push({
+			prompt: e,
+			tools: t
+		});
+	}
+	next() {
+		let e = this.turns[Math.min(this.cursor, this.turns.length - 1)];
+		return this.cursor += 1, e;
+	}
+	reset() {
+		this.cursor = 0, this.calls.length = 0;
+	}
+}, g = (e) => (e.toolCalls ?? []).map((e, t) => ({
+	type: "tool-call",
+	toolCallId: `mock_tool_${t}`,
+	toolName: e.name,
+	input: JSON.stringify(e.input ?? {})
+})), _ = (e) => (e.sources ?? []).map((e, t) => e.sourceType === "document" ? {
+	type: "source",
+	sourceType: "document",
+	id: e.id ?? `mock_src_${t}`,
+	mediaType: e.mediaType ?? "application/pdf",
+	title: e.title ?? "document",
+	...e.filename === void 0 ? {} : { filename: e.filename }
+} : {
+	type: "source",
+	sourceType: "url",
+	id: e.id ?? `mock_src_${t}`,
+	url: e.url ?? "https://example.com",
+	...e.title === void 0 ? {} : { title: e.title }
+}), v = (e) => (e.files ?? []).map((e) => ({
+	type: "file",
+	mediaType: e.mediaType,
+	data: {
+		type: "data",
+		data: e.data
+	}
+})), y = (e) => [
+	{
+		type: "text-start",
+		id: "0"
+	},
+	...e.split(/(\s+)/).filter((e) => e.length > 0).map((e) => ({
+		type: "text-delta",
+		id: "0",
+		delta: e
+	})),
+	{
+		type: "text-end",
+		id: "0"
+	}
+], b = (e) => e.length === 0 ? [] : [
+	{
+		type: "reasoning-start",
+		id: "r0"
+	},
+	{
+		type: "reasoning-delta",
+		id: "r0",
+		delta: e
+	},
+	{
+		type: "reasoning-end",
+		id: "r0"
+	}
+], x = (e, t, n, r) => ({
+	specificationVersion: "v4",
+	provider: "mock",
+	modelId: e,
+	supportedUrls: {},
+	doGenerate: async ({ prompt: i, tools: a }) => {
+		n?.record(i, (a ?? []).map((e) => e.name ?? ""));
+		let o = n?.next();
+		if (o?.error) throw Error(o.error.message);
+		let s = JSON.stringify(i).slice(0, 80), c = o?.toolCalls ?? [];
+		return c.length > 0 ? {
+			content: c.map((e, t) => ({
+				type: "tool-call",
+				toolCallId: `mock_tool_${t}`,
+				toolName: e.name,
+				input: JSON.stringify(e.input ?? {})
+			})),
+			finishReason: m("tool-calls"),
+			usage: p,
+			warnings: []
+		} : {
+			content: [{
+				type: "text",
+				text: o?.text ?? t ?? `[mock(${e})] ${s}`
+			}],
+			finishReason: m("stop"),
+			usage: p,
+			warnings: [],
+			...r === void 0 ? {} : { providerMetadata: r }
+		};
+	},
+	doStream: async ({ prompt: e, tools: r }) => {
+		n?.record(e, (r ?? []).map((e) => e.name ?? ""));
+		let i = n?.next(), a = (i?.toolCalls?.length ?? 0) > 0, o = i?.text ?? (n ? "" : t ?? "[mock] streamed response"), s = [{
+			type: "stream-start",
+			warnings: []
+		}];
+		return i?.error ? (s.push({
+			type: "error",
+			error: Error(i.error.message)
+		}), s.push({
+			type: "finish",
+			finishReason: m("error"),
+			usage: p
+		})) : (s.push(...b(i?.reasoning ?? "")), o.length > 0 && s.push(...y(o)), s.push(..._(i ?? {})), s.push(...v(i ?? {})), s.push(...g(i ?? {})), s.push({
+			type: "finish",
+			finishReason: m(a ? "tool-calls" : "stop"),
+			usage: p
+		})), { stream: new ReadableStream({ start(e) {
+			for (let t of s) e.enqueue(t);
+			e.close();
+		} }) };
+	}
+}), S = "iVBORw0KGgo=", C = "SUQzBAAAAAAAAA==", w = "video/mp4", T = "mock transcript", E = (e, t) => ({
+	specificationVersion: "v4",
+	provider: "mock",
+	modelId: e,
+	maxImagesPerCall: 8,
+	doGenerate: async ({ n }) => ({
+		images: Array.from({ length: Math.max(1, n) }, () => t),
+		warnings: [],
+		response: {
+			timestamp: /* @__PURE__ */ new Date(),
+			modelId: e,
+			headers: void 0
+		},
+		usage: {
+			inputTokens: 0,
+			outputTokens: 0,
+			totalTokens: 0
+		}
+	})
+}), D = (e, t, n) => ({
+	specificationVersion: "v4",
+	provider: "mock",
+	modelId: e,
+	doGenerate: async () => ({
+		audio: t,
+		warnings: [],
+		response: {
+			timestamp: /* @__PURE__ */ new Date(),
+			modelId: e
+		},
+		...n === void 0 ? {} : { providerMetadata: n }
+	})
+}), O = (e, t, n) => ({
+	specificationVersion: "v4",
+	provider: "mock",
+	modelId: e,
+	doGenerate: async () => ({
+		text: t,
+		segments: [],
+		language: "en",
+		durationInSeconds: 1,
+		warnings: [],
+		response: {
+			timestamp: /* @__PURE__ */ new Date(),
+			modelId: e,
+			headers: {},
+			body: void 0
+		},
+		...n === void 0 ? {} : { providerMetadata: n }
+	})
+}), k = (e, t, n) => ({
+	specificationVersion: "v4",
+	provider: "mock",
+	modelId: e,
+	maxVideosPerCall: 1,
+	doGenerate: async () => ({
+		videos: [t],
+		warnings: [],
+		response: {
+			timestamp: /* @__PURE__ */ new Date(),
+			modelId: e,
+			headers: void 0
+		},
+		...n === void 0 ? {} : { providerMetadata: n }
+	})
+}), A = (e) => e.apiKey !== void 0 || e.baseURL !== void 0, j = (e) => ({
+	...e.apiKey === void 0 ? {} : { apiKey: e.apiKey },
+	...e.baseURL === void 0 ? {} : { baseURL: e.baseURL }
+}), M = (e, t) => {
+	let r = (() => {
+		switch (e.name) {
+			case "mock": return x(e.model ?? "mock", e.mockText, e.script, e.providerMetadata);
+			case "anthropic": return A(e) ? c(j(e))(e.model) : s(e.model);
+			case "openai": return A(e) ? l(j(e))(e.model) : u(e.model);
+			case "gateway": return A(e) ? d(j(e))(e.model) : f(e.model);
+		}
+	})();
+	return n(r, t);
+}, N = (e) => {
+	switch (e.name) {
+		case "mock": return E(e.model ?? "mock", e.mockMedia?.imageBase64 ?? S);
+		case "openai": return A(e) ? l(j(e)).image(e.model) : u.image(e.model);
+		case "gateway": return i(r(j(e), o(e.maxResponseBytes)), e.model);
+		case "anthropic": throw Error("Provider 'anthropic' does not support image generation. Use 'openai' or 'gateway'.");
+	}
+}, P = (e) => {
+	switch (e.name) {
+		case "mock": return D(e.model ?? "mock", e.mockMedia?.audioBase64 ?? C, e.providerMetadata);
+		case "openai": return A(e) ? l(j(e)).speech(e.model) : u.speech(e.model);
+		case "gateway": return d(r(j(e), o(e.maxResponseBytes))).speech(e.model);
+		case "anthropic": throw Error("Provider 'anthropic' does not support speech generation. Use 'openai' or 'gateway'.");
+	}
+}, F = (e) => {
+	switch (e.name) {
+		case "mock": return O(e.model ?? "mock", e.mockMedia?.transcript ?? T, e.providerMetadata);
+		case "openai": return A(e) ? l(j(e)).transcription(e.model) : u.transcription(e.model);
+		case "gateway": return d(r(j(e), o(e.maxResponseBytes))).transcription(e.model);
+		case "anthropic": throw Error("Provider 'anthropic' does not support transcription. Use 'openai' or 'gateway'.");
+	}
+}, I = (e) => {
+	switch (e.name) {
+		case "mock": {
+			let t = e.mockMedia, n = t?.videoBase64 === void 0 ? {
+				type: "url",
+				url: t?.videoUrl ?? `https://mock.local/${e.model ?? "mock"}.mp4`,
+				mediaType: w
+			} : {
+				type: "base64",
+				data: t.videoBase64,
+				mediaType: w
+			};
+			return k(e.model ?? "mock", n, e.providerMetadata);
+		}
+		case "gateway": return a(r(j(e), o(e.maxResponseBytes)), e.model);
+		case "openai": throw Error("Provider 'openai' does not support video generation. Use 'gateway'.");
+		case "anthropic": throw Error("Provider 'anthropic' does not support video generation. Use 'gateway'.");
+	}
+}, L = null, R = (e) => {
+	let t = L;
+	return L = e, t;
+}, z = () => L, B = (e) => {
+	switch (e.toLowerCase()) {
+		case "anthropic": return "ANTHROPIC_API_KEY";
+		case "openai": return "OPENAI_API_KEY";
+		case "gateway": return "AI_GATEWAY_API_KEY";
+		default: return;
+	}
+}, V = (t = process.env) => {
+	if (L) return {
+		name: L.name,
+		chosen: !0
+	};
+	let n = e("AI_PROVIDER", t);
+	return n === void 0 ? {
+		name: "mock",
+		chosen: !1
+	} : {
+		name: n.toLowerCase(),
+		chosen: !0
+	};
+}, H = (n = process.env) => {
+	if (L) return L;
+	let r = V(n);
+	if (!r.chosen && t(n)) throw Error("No AI provider is chosen in this production process: AI_PROVIDER is unset and this call passes no `provider` (or its agent declares no `model`), so it would answer from the mock provider. Set AI_PROVIDER (anthropic | openai | gateway) for every process that makes model calls, or declare it in `defineEnv` with a `default`, or pass `provider` to the call. Set AI_PROVIDER=mock to serve mock answers deliberately.");
+	let i = r.name, a = e("AI_MODEL", n);
+	switch (i) {
+		case "mock": return {
+			name: "mock",
+			model: a ?? "mock"
+		};
+		case "anthropic": return {
+			name: "anthropic",
+			model: a ?? "claude-opus-4-8"
+		};
+		case "openai": return {
+			name: "openai",
+			model: a ?? "gpt-5.5"
+		};
+		case "gateway": {
+			let e = a === void 0 ? -1 : a.indexOf("/");
+			if (a === void 0 || e <= 0 || e === a.length - 1) throw Error("AI_PROVIDER=gateway requires AI_MODEL set to a 'creator/model' id, e.g. 'openai/gpt-5.5'");
+			return {
+				name: "gateway",
+				model: a
+			};
+		}
+		default: throw Error(`Unknown AI_PROVIDER '${i}'. Expected one of: mock, anthropic, openai, gateway.`);
+	}
+};
+//#endregion
+export { h as MockScript, z as getProviderOverride, V as providerChoice, H as providerFromEnv, B as providerKeyEnvVar, N as resolveImageModel, M as resolveModel, P as resolveSpeechModel, F as resolveTranscriptionModel, I as resolveVideoModel, R as setProviderOverride };

package/dist/realtimeEvent.js

@@ -0,0 +1,97 @@
+import { Schema as e } from "effect";
+//#region src/realtimeEvent.ts
+var t = e.String.pipe(e.check(e.isPattern(/^(?:[A-Za-z0-9+/]{4})*(?:[A-Za-z0-9+/]{2}==|[A-Za-z0-9+/]{3}=)?$/))), n = e.Number.pipe(e.check(e.isFinite()), e.check(e.isGreaterThanOrEqualTo(0))), r = e.Union([
+	e.Struct({
+		type: e.Literal("input-audio-append"),
+		audio: t
+	}),
+	e.Struct({ type: e.Literals([
+		"input-audio-commit",
+		"input-audio-clear",
+		"response-create",
+		"response-cancel"
+	]) }),
+	e.Struct({
+		type: e.Literal("conversation-item-create"),
+		item: e.Struct({
+			type: e.Literal("text-message"),
+			role: e.Literal("user"),
+			text: e.String
+		})
+	}),
+	e.Struct({
+		type: e.Literal("conversation-item-truncate"),
+		itemId: e.String,
+		contentIndex: n.pipe(e.check(e.isInt())),
+		audioEndMs: n
+	})
+]), i = { itemId: e.String }, a = {
+	responseId: e.String,
+	...i
+}, o = e.Union([
+	e.Struct({ type: e.Literal("ready") }),
+	e.Struct({
+		type: e.Literal("session-created"),
+		sessionId: e.optional(e.String)
+	}),
+	e.Struct({ type: e.Literal("session-updated") }),
+	e.Struct({
+		type: e.Literals(["speech-started", "speech-stopped"]),
+		itemId: e.optional(e.String)
+	}),
+	e.Struct({
+		type: e.Literal("audio-committed"),
+		itemId: e.optional(e.String),
+		previousItemId: e.optional(e.String)
+	}),
+	e.Struct({
+		type: e.Literal("conversation-item-added"),
+		...i
+	}),
+	e.Struct({
+		type: e.Literal("input-transcription-completed"),
+		...i,
+		transcript: e.String
+	}),
+	e.Struct({
+		type: e.Literal("response-created"),
+		responseId: e.String
+	}),
+	e.Struct({
+		type: e.Literal("response-done"),
+		responseId: e.String,
+		status: e.String
+	}),
+	e.Struct({
+		type: e.Literals([
+			"output-item-added",
+			"output-item-done",
+			"content-part-added",
+			"content-part-done",
+			"audio-done"
+		]),
+		...a
+	}),
+	e.Struct({
+		type: e.Literal("audio-delta"),
+		...a,
+		delta: t
+	}),
+	e.Struct({
+		type: e.Literals(["text-delta", "audio-transcript-delta"]),
+		...a,
+		delta: e.String
+	}),
+	e.Struct({
+		type: e.Literal("text-done"),
+		...a,
+		text: e.optional(e.String)
+	}),
+	e.Struct({
+		type: e.Literal("audio-transcript-done"),
+		...a,
+		transcript: e.optional(e.String)
+	})
+]);
+//#endregion
+export { r as RealtimeClientEvent, o as RealtimeServerEvent };