@voltro/ai 0.33.0 → 0.34.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +1801 -0
- package/dist/index.d.ts +711 -5
- package/dist/index.js +696 -243
- package/dist/inferenceQueue-uj2kKyqp.js +568 -0
- package/dist/workflow.d.ts +155 -4
- package/dist/workflow.js +171 -96
- package/package.json +5 -5
- package/dist/inferenceQueue-B25RTJZ4.js +0 -447
|
@@ -0,0 +1,568 @@
|
|
|
1
|
+
import { a as e, r as t } from "./provider-BOC5wBxx.js";
|
|
2
|
+
import { Output as n, generateObject as r, generateText as i, jsonSchema as a, stepCountIs as o, tool as s } from "ai";
|
|
3
|
+
import { Context as c, Effect as l, JSONSchema as u, Metric as d, MetricBoundaries as ee, Runtime as f, Schema as p, Stream as m } from "effect";
|
|
4
|
+
import { and as h, eq as g, generateId as _, gte as te, id as v, integer as y, json as b, queryFor as x, registerRetention as S, retentionTtlMsFromEnv as C, table as w, text as T, timestamp as E } from "@voltro/database";
|
|
5
|
+
import { createHash as D } from "node:crypto";
|
|
6
|
+
//#region src/tools.ts
|
|
7
|
+
var ne = (e, t) => e(t), re = Symbol.for("@voltro/ai/Tool"), ie = (e) => ({
|
|
8
|
+
...e,
|
|
9
|
+
[re]: !0
|
|
10
|
+
}), ae = (e) => typeof e == "object" && !!e && e[re] === !0, oe = (e) => {
|
|
11
|
+
let t = e.default;
|
|
12
|
+
if (typeof t != "function") return 0;
|
|
13
|
+
let n = 0;
|
|
14
|
+
for (let r of Object.values(e)) ae(r) && r.execute === void 0 && r.handler === void 0 && (r.handler = t, n += 1);
|
|
15
|
+
return n;
|
|
16
|
+
}, se = (e, t) => {
|
|
17
|
+
let n = t ?? f.defaultRuntime;
|
|
18
|
+
return f.runPromise(n)(e);
|
|
19
|
+
}, ce = (e, t, n) => n ? e.pipe(l.timeoutFail({
|
|
20
|
+
duration: `${n} millis`,
|
|
21
|
+
onTimeout: () => /* @__PURE__ */ Error(`tool '${t}' timed out`)
|
|
22
|
+
})) : e, le = (e, t, n) => {
|
|
23
|
+
if (e.execute) return se(ce(ne(e.execute, t), e.name, e.timeoutMs), n.runtime);
|
|
24
|
+
if (e.handler) {
|
|
25
|
+
let r = e.handler, i = r(t, n);
|
|
26
|
+
return l.isEffect(i) ? se(ce(i, e.name, e.timeoutMs), n.runtime) : i;
|
|
27
|
+
}
|
|
28
|
+
return Promise.reject(/* @__PURE__ */ Error(`tool '${e.name}' has neither execute nor handler`));
|
|
29
|
+
}, ue = (e) => {
|
|
30
|
+
let t = u.make(e);
|
|
31
|
+
return t && t.type === "object" ? t : {
|
|
32
|
+
type: "object",
|
|
33
|
+
properties: {},
|
|
34
|
+
additionalProperties: !1
|
|
35
|
+
};
|
|
36
|
+
}, de = (e) => {
|
|
37
|
+
let t = e.inputJsonSchema;
|
|
38
|
+
return t === void 0 ? ue(e.input) : t.type === "object" ? t : {
|
|
39
|
+
type: "object",
|
|
40
|
+
properties: {},
|
|
41
|
+
additionalProperties: !0
|
|
42
|
+
};
|
|
43
|
+
}, fe = (e, t = { store: void 0 }) => {
|
|
44
|
+
let n = {};
|
|
45
|
+
for (let [r, i] of Object.entries(e)) n[i.name || r] = s({
|
|
46
|
+
description: i.description,
|
|
47
|
+
inputSchema: a(de(i)),
|
|
48
|
+
execute: async (e) => {
|
|
49
|
+
let n = await l.runPromise(p.decodeUnknown(i.input)(e)), r = await le(i, n, t);
|
|
50
|
+
return i.output ? l.runPromise(p.decodeUnknown(i.output)(r)) : r;
|
|
51
|
+
}
|
|
52
|
+
});
|
|
53
|
+
return n;
|
|
54
|
+
}, O = {
|
|
55
|
+
"claude-opus-4-8": {
|
|
56
|
+
inputPer1M: 15,
|
|
57
|
+
outputPer1M: 75
|
|
58
|
+
},
|
|
59
|
+
"claude-opus-4-6": {
|
|
60
|
+
inputPer1M: 15,
|
|
61
|
+
outputPer1M: 75
|
|
62
|
+
},
|
|
63
|
+
"claude-sonnet-4-6": {
|
|
64
|
+
inputPer1M: 3,
|
|
65
|
+
outputPer1M: 15
|
|
66
|
+
},
|
|
67
|
+
"claude-haiku-4-5": {
|
|
68
|
+
inputPer1M: 1,
|
|
69
|
+
outputPer1M: 5
|
|
70
|
+
},
|
|
71
|
+
"claude-haiku-4-5-20251001": {
|
|
72
|
+
inputPer1M: 1,
|
|
73
|
+
outputPer1M: 5
|
|
74
|
+
},
|
|
75
|
+
"gpt-5.5": {
|
|
76
|
+
inputPer1M: 5,
|
|
77
|
+
outputPer1M: 30
|
|
78
|
+
},
|
|
79
|
+
"gpt-4o": {
|
|
80
|
+
inputPer1M: 2.5,
|
|
81
|
+
outputPer1M: 10
|
|
82
|
+
},
|
|
83
|
+
"gpt-4o-mini": {
|
|
84
|
+
inputPer1M: .15,
|
|
85
|
+
outputPer1M: .6
|
|
86
|
+
},
|
|
87
|
+
"gemini-2.5-pro": {
|
|
88
|
+
inputPer1M: 1.25,
|
|
89
|
+
outputPer1M: 10
|
|
90
|
+
},
|
|
91
|
+
"gemini-2.5-flash": {
|
|
92
|
+
inputPer1M: .3,
|
|
93
|
+
outputPer1M: 2.5
|
|
94
|
+
},
|
|
95
|
+
"gemini-2.0-flash": {
|
|
96
|
+
inputPer1M: .1,
|
|
97
|
+
outputPer1M: .4
|
|
98
|
+
}
|
|
99
|
+
}, k = {}, pe = (e) => {
|
|
100
|
+
let t = k;
|
|
101
|
+
return k = { ...e }, t;
|
|
102
|
+
}, A = () => ({
|
|
103
|
+
...O,
|
|
104
|
+
...k
|
|
105
|
+
}), j = (e) => {
|
|
106
|
+
let t = A();
|
|
107
|
+
if (t[e] !== void 0) return t[e];
|
|
108
|
+
let n = e.indexOf("/");
|
|
109
|
+
if (n > 0) {
|
|
110
|
+
let r = e.slice(n + 1);
|
|
111
|
+
if (t[r] !== void 0) return t[r];
|
|
112
|
+
}
|
|
113
|
+
}, M = (e, t) => {
|
|
114
|
+
let n = e.inputTokens ?? 0, r = e.outputTokens ?? 0;
|
|
115
|
+
if (t.actualCostUsd !== void 0) {
|
|
116
|
+
let e = n + r, i = e > 0 ? n / e : 0;
|
|
117
|
+
return {
|
|
118
|
+
inputTokens: n,
|
|
119
|
+
outputTokens: r,
|
|
120
|
+
inputCostUsd: t.actualCostUsd * i,
|
|
121
|
+
outputCostUsd: t.actualCostUsd * (1 - i),
|
|
122
|
+
totalCostUsd: t.actualCostUsd,
|
|
123
|
+
costSource: "gateway"
|
|
124
|
+
};
|
|
125
|
+
}
|
|
126
|
+
let i = t.price ?? j(t.model) ?? {
|
|
127
|
+
inputPer1M: 0,
|
|
128
|
+
outputPer1M: 0
|
|
129
|
+
}, a = n / 1e6 * i.inputPer1M, o = r / 1e6 * i.outputPer1M;
|
|
130
|
+
return {
|
|
131
|
+
inputTokens: n,
|
|
132
|
+
outputTokens: r,
|
|
133
|
+
inputCostUsd: a,
|
|
134
|
+
outputCostUsd: o,
|
|
135
|
+
totalCostUsd: a + o,
|
|
136
|
+
costSource: "estimated"
|
|
137
|
+
};
|
|
138
|
+
}, N = "_voltro_ai_usage", P = w(N, {
|
|
139
|
+
id: v({ prefix: "aiu" }),
|
|
140
|
+
tenantId: T().nullable(),
|
|
141
|
+
provider: T(),
|
|
142
|
+
model: T(),
|
|
143
|
+
operation: T(),
|
|
144
|
+
agent: T().nullable(),
|
|
145
|
+
inputTokens: y().default(0),
|
|
146
|
+
outputTokens: y().default(0),
|
|
147
|
+
costMicroUsd: y().default(0),
|
|
148
|
+
costSource: T().default("estimated"),
|
|
149
|
+
promptId: T().nullable(),
|
|
150
|
+
promptDigest: T().nullable(),
|
|
151
|
+
promptRevision: y().nullable(),
|
|
152
|
+
calledAt: E().default("now")
|
|
153
|
+
}).index(["tenantId"]).index(["promptDigest"]), me = (e) => {
|
|
154
|
+
let t = e?.gateway?.cost, n = typeof t == "number" ? t : typeof t == "string" ? Number(t) : NaN;
|
|
155
|
+
return Number.isFinite(n) ? n : void 0;
|
|
156
|
+
}, he = (e, t) => (S({
|
|
157
|
+
source: "plugin",
|
|
158
|
+
table: N,
|
|
159
|
+
timeColumn: "calledAt",
|
|
160
|
+
ttlMs: C(process.env.VOLTRO_AI_USAGE_TTL_HOURS, 24 * 365)
|
|
161
|
+
}), l.promise(async () => {
|
|
162
|
+
let n = M(t.usage, {
|
|
163
|
+
model: t.model,
|
|
164
|
+
...t.price === void 0 ? {} : { price: t.price },
|
|
165
|
+
...t.actualCostUsd === void 0 ? {} : { actualCostUsd: t.actualCostUsd }
|
|
166
|
+
});
|
|
167
|
+
return await e.insert(N, {
|
|
168
|
+
tenantId: t.tenantId ?? null,
|
|
169
|
+
provider: t.provider,
|
|
170
|
+
model: t.model,
|
|
171
|
+
operation: t.operation,
|
|
172
|
+
agent: t.agent ?? null,
|
|
173
|
+
inputTokens: n.inputTokens,
|
|
174
|
+
outputTokens: n.outputTokens,
|
|
175
|
+
costMicroUsd: Math.round(n.totalCostUsd * 1e6),
|
|
176
|
+
costSource: n.costSource,
|
|
177
|
+
promptId: t.prompt?.promptId ?? null,
|
|
178
|
+
promptDigest: t.prompt?.digest ?? null,
|
|
179
|
+
promptRevision: t.prompt?.revision ?? null,
|
|
180
|
+
calledAt: t.calledAt ?? /* @__PURE__ */ new Date()
|
|
181
|
+
}), n;
|
|
182
|
+
})), ge = (e, t = {}) => l.promise(async () => {
|
|
183
|
+
let n = [
|
|
184
|
+
...t.tenantId === void 0 ? [] : [g("tenantId", t.tenantId)],
|
|
185
|
+
...t.since === void 0 ? [] : [te("calledAt", t.since)],
|
|
186
|
+
...t.promptDigest === void 0 ? [] : [g("promptDigest", t.promptDigest)]
|
|
187
|
+
], r = x(P), i = n.length === 0 ? r : r.where(n.length === 1 ? n[0] : h(...n));
|
|
188
|
+
return (await e.query(i.descriptor)).reduce((e, t) => e + (Number(t.costMicroUsd) || 0), 0) / 1e6;
|
|
189
|
+
}), F = class extends p.TaggedError()("AiBudgetExceeded", {
|
|
190
|
+
tenantId: p.NullOr(p.String),
|
|
191
|
+
limitUsd: p.Number,
|
|
192
|
+
spentUsd: p.Number,
|
|
193
|
+
attemptedUsd: p.Number
|
|
194
|
+
}) {}, I = "_voltro_ai_budget", L = w(I, {
|
|
195
|
+
id: v({ prefix: "aib" }),
|
|
196
|
+
bucket: T().maxLength(191),
|
|
197
|
+
tenantId: T().nullable(),
|
|
198
|
+
reservedMicroUsd: y().default(0),
|
|
199
|
+
updatedAt: E().default("now")
|
|
200
|
+
}).unique("byAiBudgetBucket", ["bucket"]), _e = (e, t) => `${e ?? "global"}::${t === void 0 ? "all" : String(t.getTime())}`, R = 32, ve = (e, t) => l.gen(function* () {
|
|
201
|
+
S({
|
|
202
|
+
source: "plugin",
|
|
203
|
+
table: I,
|
|
204
|
+
timeColumn: "updatedAt",
|
|
205
|
+
ttlMs: C(process.env.VOLTRO_AI_USAGE_TTL_HOURS, 24 * 365)
|
|
206
|
+
});
|
|
207
|
+
let n = t.addUsd ?? 0, r = Math.round(n * 1e6), i = Math.round(t.limitUsd * 1e6), a = _e(t.tenantId, t.since), o = t.tenantId ?? null, s = (e) => new F({
|
|
208
|
+
tenantId: o,
|
|
209
|
+
limitUsd: t.limitUsd,
|
|
210
|
+
spentUsd: e / 1e6,
|
|
211
|
+
attemptedUsd: n
|
|
212
|
+
}), c = yield* l.promise(async () => {
|
|
213
|
+
for (let t = 0; t < R; t++) {
|
|
214
|
+
t > 0 && await new Promise((e) => setTimeout(e, t));
|
|
215
|
+
let n = (await e.query(x(L).where(g("bucket", a)).descriptor))[0], s = n && Number(n.reservedMicroUsd) || 0;
|
|
216
|
+
if (s + r > i) return {
|
|
217
|
+
ok: !1,
|
|
218
|
+
reservedMicro: s
|
|
219
|
+
};
|
|
220
|
+
if (r === 0) return { ok: !0 };
|
|
221
|
+
if (n) {
|
|
222
|
+
if (await e.updateMany("_voltro_ai_budget", {
|
|
223
|
+
reservedMicroUsd: s + r,
|
|
224
|
+
updatedAt: /* @__PURE__ */ new Date()
|
|
225
|
+
}, { where: h(g("id", String(n.id)), g("reservedMicroUsd", s)) }) === 1) return { ok: !0 };
|
|
226
|
+
} else {
|
|
227
|
+
let t = _({
|
|
228
|
+
kind: "typeid",
|
|
229
|
+
prefix: "aib"
|
|
230
|
+
}, I), n = await e.insertIgnore(I, {
|
|
231
|
+
id: t,
|
|
232
|
+
bucket: a,
|
|
233
|
+
tenantId: o,
|
|
234
|
+
reservedMicroUsd: r,
|
|
235
|
+
updatedAt: /* @__PURE__ */ new Date()
|
|
236
|
+
}, { conflictColumns: ["bucket"] });
|
|
237
|
+
if (String(n.id) === t) return { ok: !0 };
|
|
238
|
+
}
|
|
239
|
+
}
|
|
240
|
+
throw Error(`requireAiBudget: contention on bucket '${a}' exceeded ${R} attempts`);
|
|
241
|
+
});
|
|
242
|
+
if (!c.ok) return yield* l.fail(s(c.reservedMicro));
|
|
243
|
+
}), ye = ee.exponential({
|
|
244
|
+
start: .001,
|
|
245
|
+
factor: 2,
|
|
246
|
+
count: 16
|
|
247
|
+
}), z = d.counter("voltro_ai_calls_total", { description: "Total AI provider calls (generate + stream)." }), B = d.counter("voltro_ai_call_errors_total", { description: "AI provider calls that errored." }), V = d.histogram("voltro_ai_call_duration_seconds", ye, "AI provider call duration in seconds."), be = d.counter("voltro_ai_input_tokens_total", {
|
|
248
|
+
description: "Total prompt (input) tokens across AI calls.",
|
|
249
|
+
incremental: !0
|
|
250
|
+
}), xe = d.counter("voltro_ai_output_tokens_total", {
|
|
251
|
+
description: "Total completion (output) tokens across AI calls.",
|
|
252
|
+
incremental: !0
|
|
253
|
+
}), Se = d.counter("voltro_ai_cost_microusd_total", {
|
|
254
|
+
description: "Total estimated AI spend in micro-USD (USD × 1e6).",
|
|
255
|
+
incremental: !0
|
|
256
|
+
}), H = (e, t) => ({
|
|
257
|
+
provider: e.name,
|
|
258
|
+
model: e.model ?? e.name,
|
|
259
|
+
operation: t
|
|
260
|
+
}), U = (e, t, n) => e.pipe(l.tagMetrics("provider", t.provider), l.tagMetrics("model", t.model), l.tagMetrics("operation", t.operation), l.tagMetrics("status", n)), Ce = (e, t) => {
|
|
261
|
+
let n = t.inputTokens ?? 0, r = t.outputTokens ?? 0, i = M({
|
|
262
|
+
inputTokens: n,
|
|
263
|
+
outputTokens: r
|
|
264
|
+
}, { model: e.model }), a = Math.round(i.totalCostUsd * 1e6);
|
|
265
|
+
return l.all([
|
|
266
|
+
U(d.incrementBy(be, n), e, "ok"),
|
|
267
|
+
U(d.incrementBy(xe, r), e, "ok"),
|
|
268
|
+
U(d.incrementBy(Se, a), e, "ok")
|
|
269
|
+
], { discard: !0 });
|
|
270
|
+
}, W = (e, t, n) => l.gen(function* () {
|
|
271
|
+
let r = yield* l.clockWith((e) => e.currentTimeMillis);
|
|
272
|
+
return yield* n.pipe(l.tap((n) => Ce(e, t(n))), l.tap(() => U(d.increment(z), e, "ok").pipe(l.zipRight(l.clockWith((e) => e.currentTimeMillis).pipe(l.flatMap((t) => U(d.update(V, (t - r) / 1e3), e, "ok")))))), l.tapError(() => U(d.increment(z), e, "error").pipe(l.zipRight(U(d.increment(B), e, "error")), l.zipRight(l.clockWith((e) => e.currentTimeMillis).pipe(l.flatMap((t) => U(d.update(V, (t - r) / 1e3), e, "error")))))), l.withSpan("voltro.ai.call", { attributes: {
|
|
273
|
+
"ai.provider": e.provider,
|
|
274
|
+
"ai.model": e.model,
|
|
275
|
+
"ai.operation": e.operation
|
|
276
|
+
} }));
|
|
277
|
+
}), we = (e, t) => {
|
|
278
|
+
let n = Date.now();
|
|
279
|
+
return t.pipe(m.tap((t) => t._tag === "done" ? U(d.increment(z), e, "ok").pipe(l.zipRight(U(d.update(V, (Date.now() - n) / 1e3), e, "ok")), l.zipRight(Ce(e, t.usage ?? {}))) : t._tag === "error" ? U(d.increment(z), e, "error").pipe(l.zipRight(U(d.increment(B), e, "error")), l.zipRight(U(d.update(V, (Date.now() - n) / 1e3), e, "error"))) : l.void), m.withSpan("voltro.ai.call", { attributes: {
|
|
280
|
+
"ai.provider": e.provider,
|
|
281
|
+
"ai.model": e.model,
|
|
282
|
+
"ai.operation": e.operation
|
|
283
|
+
} }));
|
|
284
|
+
}, G = class extends p.TaggedError()("AiError", {
|
|
285
|
+
reason: p.Literal("generation", "decode"),
|
|
286
|
+
message: p.String
|
|
287
|
+
}) {}, K = (e, t, n) => {
|
|
288
|
+
let r = [e, ...t ?? []], i = (e) => n(r[e]).pipe(l.catchTag("AiError", (t) => t.reason === "generation" && e + 1 < r.length ? i(e + 1) : l.fail(t)));
|
|
289
|
+
return i(0);
|
|
290
|
+
}, Te = (n) => K(n.provider ?? t(), n.fallbacks, (t) => W(H(t, "generateText"), (e) => e.usage, l.tryPromise({
|
|
291
|
+
try: async () => {
|
|
292
|
+
let r = await i({
|
|
293
|
+
model: e(t),
|
|
294
|
+
prompt: n.prompt,
|
|
295
|
+
...n.system === void 0 ? {} : { system: n.system },
|
|
296
|
+
...n.maxTokens === void 0 ? {} : { maxOutputTokens: n.maxTokens },
|
|
297
|
+
...n.providerOptions === void 0 ? {} : { providerOptions: n.providerOptions }
|
|
298
|
+
});
|
|
299
|
+
return {
|
|
300
|
+
text: r.text,
|
|
301
|
+
usage: {
|
|
302
|
+
inputTokens: r.usage.inputTokens,
|
|
303
|
+
outputTokens: r.usage.outputTokens,
|
|
304
|
+
totalTokens: r.usage.totalTokens
|
|
305
|
+
},
|
|
306
|
+
providerMetadata: r.providerMetadata
|
|
307
|
+
};
|
|
308
|
+
},
|
|
309
|
+
catch: (e) => new G({
|
|
310
|
+
reason: "generation",
|
|
311
|
+
message: String(e?.message ?? e)
|
|
312
|
+
})
|
|
313
|
+
}))), q = (e, t) => l.gen(function* () {
|
|
314
|
+
let n = u.make(e);
|
|
315
|
+
if (n?.type !== "object") {
|
|
316
|
+
let e = n?.anyOf === void 0 ? `type:${JSON.stringify(n?.type ?? null)}` : "anyOf (e.g. Schema.Struct({}) or a union)";
|
|
317
|
+
return yield* l.fail(new G({
|
|
318
|
+
reason: "generation",
|
|
319
|
+
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(...) })).`
|
|
320
|
+
}));
|
|
321
|
+
}
|
|
322
|
+
return n;
|
|
323
|
+
}), Ee = (e, t) => q(e, t), De = (n) => K(n.provider ?? t(), n.fallbacks, (t) => W(H(t, "generateObject"), (e) => e.usage, l.gen(function* () {
|
|
324
|
+
let i = e(t), o = yield* l.tryPromise({
|
|
325
|
+
try: () => r({
|
|
326
|
+
model: i,
|
|
327
|
+
schema: a(n.jsonSchema),
|
|
328
|
+
prompt: n.prompt,
|
|
329
|
+
...n.system === void 0 ? {} : { system: n.system },
|
|
330
|
+
...n.maxTokens === void 0 ? {} : { maxOutputTokens: n.maxTokens },
|
|
331
|
+
...n.providerOptions === void 0 ? {} : { providerOptions: n.providerOptions }
|
|
332
|
+
}),
|
|
333
|
+
catch: (e) => new G({
|
|
334
|
+
reason: "generation",
|
|
335
|
+
message: String(e?.message ?? e)
|
|
336
|
+
})
|
|
337
|
+
});
|
|
338
|
+
return {
|
|
339
|
+
object: o.object,
|
|
340
|
+
usage: {
|
|
341
|
+
inputTokens: o.usage.inputTokens,
|
|
342
|
+
outputTokens: o.usage.outputTokens,
|
|
343
|
+
totalTokens: o.usage.totalTokens
|
|
344
|
+
},
|
|
345
|
+
providerMetadata: o.providerMetadata
|
|
346
|
+
};
|
|
347
|
+
}))), Oe = (e) => l.gen(function* () {
|
|
348
|
+
let t = yield* De({
|
|
349
|
+
jsonSchema: yield* q(e.schema, "generateObject"),
|
|
350
|
+
prompt: e.prompt,
|
|
351
|
+
...e.system === void 0 ? {} : { system: e.system },
|
|
352
|
+
...e.provider === void 0 ? {} : { provider: e.provider },
|
|
353
|
+
...e.fallbacks === void 0 ? {} : { fallbacks: e.fallbacks },
|
|
354
|
+
...e.maxTokens === void 0 ? {} : { maxTokens: e.maxTokens },
|
|
355
|
+
...e.providerOptions === void 0 ? {} : { providerOptions: e.providerOptions }
|
|
356
|
+
}), n = yield* p.decodeUnknown(e.schema)(t.object).pipe(l.mapError((e) => new G({
|
|
357
|
+
reason: "decode",
|
|
358
|
+
message: `model output failed schema decode: ${String(e)}`
|
|
359
|
+
})));
|
|
360
|
+
return {
|
|
361
|
+
...t,
|
|
362
|
+
object: n
|
|
363
|
+
};
|
|
364
|
+
}), ke = (r) => l.gen(function* () {
|
|
365
|
+
let s = yield* l.runtime(), c = yield* q(r.schema, "generateObjectWithTools"), u = {
|
|
366
|
+
store: void 0,
|
|
367
|
+
...r.toolContext,
|
|
368
|
+
runtime: s
|
|
369
|
+
}, d = fe(r.tools, u);
|
|
370
|
+
return yield* K(r.provider ?? t(), r.fallbacks, (t) => W(H(t, "generateObjectWithTools"), (e) => e.usage, l.gen(function* () {
|
|
371
|
+
let s = e(t), u = yield* l.tryPromise({
|
|
372
|
+
try: () => i({
|
|
373
|
+
model: s,
|
|
374
|
+
prompt: r.prompt,
|
|
375
|
+
tools: d,
|
|
376
|
+
stopWhen: o(r.maxSteps ?? 8),
|
|
377
|
+
output: n.object({ schema: a(c) }),
|
|
378
|
+
...r.system === void 0 ? {} : { system: r.system },
|
|
379
|
+
...r.maxTokens === void 0 ? {} : { maxOutputTokens: r.maxTokens },
|
|
380
|
+
...r.providerOptions === void 0 ? {} : { providerOptions: r.providerOptions }
|
|
381
|
+
}),
|
|
382
|
+
catch: (e) => new G({
|
|
383
|
+
reason: "generation",
|
|
384
|
+
message: String(e?.message ?? e)
|
|
385
|
+
})
|
|
386
|
+
});
|
|
387
|
+
return {
|
|
388
|
+
object: yield* p.decodeUnknown(r.schema)(u.output).pipe(l.mapError((e) => new G({
|
|
389
|
+
reason: "decode",
|
|
390
|
+
message: `model output failed schema decode: ${String(e)}`
|
|
391
|
+
}))),
|
|
392
|
+
usage: {
|
|
393
|
+
inputTokens: u.usage.inputTokens,
|
|
394
|
+
outputTokens: u.usage.outputTokens,
|
|
395
|
+
totalTokens: u.usage.totalTokens
|
|
396
|
+
},
|
|
397
|
+
providerMetadata: u.providerMetadata
|
|
398
|
+
};
|
|
399
|
+
})));
|
|
400
|
+
}), Ae = class extends c.Tag("@voltro/AiService")() {}, je = {
|
|
401
|
+
generateText: (e) => Te(e),
|
|
402
|
+
generateObject: (e) => Oe(e)
|
|
403
|
+
}, Me = (e) => `sha256:${D("sha256").update(e).digest("hex").slice(0, 16)}`, Ne = class extends p.TaggedError()("PromptRenderError", {
|
|
404
|
+
promptId: p.String,
|
|
405
|
+
missing: p.Array(p.String),
|
|
406
|
+
message: p.String
|
|
407
|
+
}) {}, Pe = /\{\{\s*([A-Za-z0-9_.]+)\s*\}\}/g, Fe = (e) => {
|
|
408
|
+
let t = /* @__PURE__ */ new Set();
|
|
409
|
+
for (let n of e.matchAll(Pe)) {
|
|
410
|
+
let e = n[1];
|
|
411
|
+
e !== void 0 && t.add(e);
|
|
412
|
+
}
|
|
413
|
+
return [...t];
|
|
414
|
+
}, Ie = (e, t) => Me(`${e.length}:${e}${(t ?? "").length}:${t ?? ""}`), Le = (e) => {
|
|
415
|
+
let t = Ie(e.template, e.system), n = Fe(e.template);
|
|
416
|
+
return {
|
|
417
|
+
...e,
|
|
418
|
+
digest: t,
|
|
419
|
+
variables: n,
|
|
420
|
+
render: (r = {}) => l.suspend(() => {
|
|
421
|
+
let i = n.filter((e) => r[e] === void 0);
|
|
422
|
+
if (i.length > 0) return l.fail(new Ne({
|
|
423
|
+
promptId: e.id,
|
|
424
|
+
missing: i,
|
|
425
|
+
message: `prompt '${e.id}' needs ${i.map((e) => `{{${e}}}`).join(", ")}, which ${i.length === 1 ? "was" : "were"} not supplied.`
|
|
426
|
+
}));
|
|
427
|
+
let a = e.template.replace(Pe, (e, t) => String(r[t]));
|
|
428
|
+
return l.succeed({
|
|
429
|
+
promptId: e.id,
|
|
430
|
+
digest: t,
|
|
431
|
+
...e.label === void 0 ? {} : { label: e.label },
|
|
432
|
+
prompt: a,
|
|
433
|
+
...e.system === void 0 ? {} : { system: e.system },
|
|
434
|
+
definition: e
|
|
435
|
+
});
|
|
436
|
+
})
|
|
437
|
+
};
|
|
438
|
+
}, Re = (e) => typeof e == "object" && !!e && typeof e.prompt == "string" && typeof e.digest == "string" && typeof e.promptId == "string", J = "_voltro_prompts", Y = w(J, {
|
|
439
|
+
id: v({ prefix: "prm" }),
|
|
440
|
+
promptId: T().maxLength(191),
|
|
441
|
+
digest: T().maxLength(191),
|
|
442
|
+
revision: y().default(1),
|
|
443
|
+
label: T().nullable(),
|
|
444
|
+
template: T().nullable(),
|
|
445
|
+
system: T().nullable(),
|
|
446
|
+
templateChars: y().default(0),
|
|
447
|
+
firstSeenAt: E().default("now"),
|
|
448
|
+
lastUsedAt: E().default("now")
|
|
449
|
+
}).unique("byPromptsVersion", ["promptId", "digest"]).index(["promptId"]), ze = 24 * 365, Be = () => {
|
|
450
|
+
S({
|
|
451
|
+
source: "framework",
|
|
452
|
+
table: J,
|
|
453
|
+
timeColumn: "lastUsedAt",
|
|
454
|
+
ttlMs: C(process.env.VOLTRO_AI_PROMPTS_TTL_HOURS, ze),
|
|
455
|
+
envVar: "VOLTRO_AI_PROMPTS_TTL_HOURS",
|
|
456
|
+
label: "ai prompt versions"
|
|
457
|
+
});
|
|
458
|
+
}, X = (e) => ({
|
|
459
|
+
promptId: String(e.promptId),
|
|
460
|
+
digest: String(e.digest),
|
|
461
|
+
revision: Number(e.revision) || 1,
|
|
462
|
+
label: e.label ?? null,
|
|
463
|
+
template: e.template ?? null,
|
|
464
|
+
system: e.system ?? null,
|
|
465
|
+
firstSeenAt: e.firstSeenAt ?? null,
|
|
466
|
+
lastUsedAt: e.lastUsedAt ?? null
|
|
467
|
+
}), Ve = (e, t) => l.gen(function* () {
|
|
468
|
+
Be();
|
|
469
|
+
let n = t.digest ?? Le(t).digest;
|
|
470
|
+
return yield* l.promise(async () => {
|
|
471
|
+
let r = (await e.query(x(Y).where(h(g("promptId", t.id), g("digest", n))).descriptor))[0];
|
|
472
|
+
if (r !== void 0) {
|
|
473
|
+
let t = /* @__PURE__ */ new Date();
|
|
474
|
+
return await e.updateMany(J, { lastUsedAt: t }, { where: g("id", String(r.id)) }), X({
|
|
475
|
+
...r,
|
|
476
|
+
lastUsedAt: t
|
|
477
|
+
});
|
|
478
|
+
}
|
|
479
|
+
let i = (await e.query(x(Y).where(g("promptId", t.id)).descriptor)).reduce((e, t) => Math.max(e, Number(t.revision) || 0), 0) + 1, a = /* @__PURE__ */ new Date(), o = {
|
|
480
|
+
id: _({
|
|
481
|
+
kind: "typeid",
|
|
482
|
+
prefix: "prm"
|
|
483
|
+
}, J),
|
|
484
|
+
promptId: t.id,
|
|
485
|
+
digest: n,
|
|
486
|
+
revision: i,
|
|
487
|
+
label: t.label ?? null,
|
|
488
|
+
template: t.template,
|
|
489
|
+
system: t.system ?? null,
|
|
490
|
+
templateChars: t.template.length,
|
|
491
|
+
firstSeenAt: a,
|
|
492
|
+
lastUsedAt: a
|
|
493
|
+
};
|
|
494
|
+
return X(await e.insertIgnore(J, o, { conflictColumns: ["promptId", "digest"] }) ?? o);
|
|
495
|
+
});
|
|
496
|
+
}), He = (e, t) => l.promise(async () => {
|
|
497
|
+
let n = (await e.query(x(Y).where(g("digest", t)).descriptor))[0];
|
|
498
|
+
return n === void 0 ? null : X(n);
|
|
499
|
+
}), Ue = (e, t) => l.promise(async () => [...await e.query(x(Y).where(g("promptId", t)).orderBy("revision", "desc").descriptor)].sort((e, t) => (Number(t.revision) || 0) - (Number(e.revision) || 0)).map(X)), Z = "_voltro_ai_inferences", We = w(Z, {
|
|
500
|
+
id: v({ prefix: "aiinf" }),
|
|
501
|
+
tenantId: T().nullable(),
|
|
502
|
+
executionId: T(),
|
|
503
|
+
workflowName: T(),
|
|
504
|
+
runId: T().nullable(),
|
|
505
|
+
signalName: T(),
|
|
506
|
+
stepName: T(),
|
|
507
|
+
status: T().oneOf([
|
|
508
|
+
"pending",
|
|
509
|
+
"running",
|
|
510
|
+
"succeeded",
|
|
511
|
+
"failed"
|
|
512
|
+
]).default("pending"),
|
|
513
|
+
kind: T().oneOf(["text", "object"]).default("text"),
|
|
514
|
+
provider: T(),
|
|
515
|
+
model: T(),
|
|
516
|
+
prompt: T(),
|
|
517
|
+
system: T().nullable(),
|
|
518
|
+
promptId: T().nullable(),
|
|
519
|
+
promptDigest: T().nullable(),
|
|
520
|
+
promptRevision: y().nullable(),
|
|
521
|
+
maxTokens: y().nullable(),
|
|
522
|
+
providerOptions: b().nullable(),
|
|
523
|
+
jsonSchema: b().nullable(),
|
|
524
|
+
result: b().nullable(),
|
|
525
|
+
errorMessage: T().nullable(),
|
|
526
|
+
attempts: y().default(0),
|
|
527
|
+
leaseUntil: E().nullable(),
|
|
528
|
+
claimedBy: T().nullable(),
|
|
529
|
+
enqueuedAt: E().default("now"),
|
|
530
|
+
startedAt: E().nullable(),
|
|
531
|
+
completedAt: E().nullable(),
|
|
532
|
+
durationMs: y().nullable()
|
|
533
|
+
}).index(["status", "enqueuedAt"]).index(["status", "leaseUntil"]).index(["executionId"]).index(["tenantId"]), Ge = 5 * 6e4, Ke = 4, Q = (e, t) => {
|
|
534
|
+
let n = typeof e == "number" ? e : Number(e);
|
|
535
|
+
return Number.isFinite(n) ? Math.trunc(n) : t;
|
|
536
|
+
}, $ = (e) => e == null ? null : String(e), qe = (e) => ({
|
|
537
|
+
id: String(e.id),
|
|
538
|
+
tenantId: $(e.tenantId),
|
|
539
|
+
executionId: String(e.executionId),
|
|
540
|
+
workflowName: String(e.workflowName),
|
|
541
|
+
runId: $(e.runId),
|
|
542
|
+
signalName: String(e.signalName),
|
|
543
|
+
stepName: String(e.stepName),
|
|
544
|
+
kind: e.kind === "object" ? "object" : "text",
|
|
545
|
+
provider: String(e.provider),
|
|
546
|
+
model: String(e.model),
|
|
547
|
+
prompt: String(e.prompt ?? ""),
|
|
548
|
+
system: $(e.system),
|
|
549
|
+
promptId: $(e.promptId),
|
|
550
|
+
promptDigest: $(e.promptDigest),
|
|
551
|
+
promptRevision: e.promptRevision === null || e.promptRevision === void 0 ? null : Q(e.promptRevision, 0),
|
|
552
|
+
maxTokens: e.maxTokens === null || e.maxTokens === void 0 ? null : Q(e.maxTokens, 0),
|
|
553
|
+
providerOptions: e.providerOptions ?? null,
|
|
554
|
+
jsonSchema: e.jsonSchema ?? null,
|
|
555
|
+
attempts: Q(e.attempts, 0)
|
|
556
|
+
}), Je = (e) => {
|
|
557
|
+
let t = e.attempts + 1;
|
|
558
|
+
return t >= e.maxAttempts ? {
|
|
559
|
+
kind: "give-up",
|
|
560
|
+
attempts: t
|
|
561
|
+
} : {
|
|
562
|
+
kind: "retry",
|
|
563
|
+
attempts: t,
|
|
564
|
+
delayMs: e.retryAfterMs !== void 0 && e.retryAfterMs > 0 ? e.retryAfterMs : Math.min(e.baseDelayMs * 2 ** (t - 1), e.maxDelayMs)
|
|
565
|
+
};
|
|
566
|
+
}, Ye = (e) => `voltro.ai.infer:${e}`, Xe = (e, t) => `aiinf_${D("sha256").update(`${e}\u0000${t}`).digest("hex").slice(0, 40)}`;
|
|
567
|
+
//#endregion
|
|
568
|
+
export { W as A, A as B, Oe as C, Ee as D, Te as E, L as F, fe as G, he as H, ge as I, oe as J, ie as K, P as L, N as M, F as N, H as O, O as P, M as R, je as S, ke as T, ve as U, j as V, pe as W, Ue as _, Je as a, G as b, qe as c, Ne as d, Le as f, Ie as g, He as h, We as i, I as j, we as k, ze as l, Me as m, Ke as n, Ye as o, Re as p, ae as q, Ge as r, Xe as s, Z as t, J as u, Y as v, De as w, Ae as x, Ve as y, me as z };
|