@allbluecn/web-app 0.12.0 → 0.12.2

package/src/server/ai-gateway/media/index.ts

@@ -0,0 +1,231 @@
+import { randomUUID } from "node:crypto"
+import { Buffer } from "node:buffer"
+import { generateImage, generateVideo, getVideoJobStatus, generateSpeech, generateTranscription } from "@tanstack/ai"
+import type { PersistedArtifactRef, ImageGenerationResult } from "@tanstack/ai"
+import { withGenerationPersistence, defineGenerationRunStore } from "@tanstack/ai-persistence"
+import type { GenerationRunRecord } from "@tanstack/ai-persistence"
+import { prisma, type Prisma } from "@allbluecn/database"
+import { z } from "zod"
+import { assertFetchableUrl } from "@/server/fetch-guard"
+import { resolveImageAdapter, resolveVideoAdapter, resolveSpeechAdapter, resolveTranscriptionAdapter, type ResolveAdapterOptions } from "./adapters.ts"
+import { recordGenerationRun } from "./recorder.ts"
+import { runChatAsAgent } from "@/server/ai-gateway/runs/agent-runtime"
+import { prototypeSchema, stripCodeFence, type GeneratePrototypeResult } from "./prototype-schema.ts"
+import type { SchemaInput } from "@tanstack/ai"
+export { recordGenerationRun } from "./recorder.ts"
+
+export interface MediaGenOptions extends ResolveAdapterOptions {
+  userId: string
+  threadId: string
+}
+
+const runSelect = {
+  id: true, userId: true, threadId: true, activity: true, provider: true, model: true,
+  status: true, startedAt: true, finishedAt: true, error: true, result: true, artifacts: true, usage: true,
+} satisfies Prisma.GenerationRunSelect
+const statusSchema = z.enum(["running", "completed", "failed", "interrupted"])
+
+async function requireThread(opts: Pick<MediaGenOptions, "userId" | "threadId">) {
+  if (!opts.userId || !opts.threadId) throw new Error("Thread not found")
+  const thread = await prisma.chatThread.findFirst({
+    where: { id: opts.threadId, userId: opts.userId }, select: { id: true },
+  })
+  if (!thread) throw new Error("Thread not found")
+}
+
+function json(value: unknown): Prisma.InputJsonValue {
+  return JSON.parse(JSON.stringify(value))
+}
+
+/** 与 SDK 非流式 generateVideo 派生 run id 对齐（videoRunIdForJob，SDK 未导出）：
+ *  `video:<encodeURIComponent(provider)>:<encodeURIComponent(jobId)>`。
+ *  本地拼法与 pollVideoForUser 的查询 key 共用同一来源，避免两份拼法漂移。 */
+function videoRunStoreKey(adapterName: string, jobId: string) {
+  return `video:${encodeURIComponent(adapterName)}:${encodeURIComponent(jobId)}`
+}
+
+export function createGenerationRunStore(opts: Pick<MediaGenOptions, "userId" | "threadId">, input?: unknown) {
+  const scope = { userId: opts.userId, threadId: opts.threadId }
+  const mapRow = (row: Prisma.GenerationRunGetPayload<{ select: typeof runSelect }>): GenerationRunRecord => ({
+    runId: row.id, threadId: row.threadId, activity: row.activity ?? "", provider: row.provider ?? "", model: row.model ?? "",
+    status: statusSchema.parse(row.status), startedAt: row.startedAt.getTime(),
+    ...(row.finishedAt ? { finishedAt: row.finishedAt.getTime() } : {}),
+    ...(row.error ? { error: { message: row.error } } : {}), result: row.result,
+  })
+  return defineGenerationRunStore({
+    async createOrResume(record) {
+      await requireThread(opts)
+      if (record.threadId !== opts.threadId) throw new Error("Thread not found")
+      const row = await prisma.generationRun.upsert({
+        where: { id: record.runId }, update: {},
+        create: { id: record.runId, ...scope, activity: record.activity, kind: record.activity,
+          provider: record.provider, model: record.model, status: record.status ?? "running",
+          startedAt: new Date(record.startedAt), ...(input === undefined ? {} : { input: json(input) }),
+        }, select: runSelect,
+      })
+      if (row.userId !== opts.userId || row.threadId !== opts.threadId) throw new Error("Run not found")
+      return mapRow(row)
+    },
+    async update(runId, patch) {
+      const data: Prisma.GenerationRunUpdateManyMutationInput = {}
+      if (patch.status !== undefined) data.status = patch.status
+      if (patch.finishedAt !== undefined) data.finishedAt = new Date(patch.finishedAt)
+      if (patch.error !== undefined) data.error = patch.error.message
+      if (patch.result !== undefined) data.result = json(patch.result)
+      if (patch.artifacts !== undefined) data.artifacts = json(patch.artifacts)
+      if (patch.usage !== undefined) data.usage = json(patch.usage)
+      await prisma.generationRun.updateMany({ where: { id: runId, ...scope }, data })
+    },
+    async get(runId) {
+      const row = await prisma.generationRun.findFirst({ where: { id: runId, ...scope }, select: runSelect })
+      return row ? mapRow(row) : null
+    },
+    async findLatestForThread(threadId) {
+      if (threadId !== opts.threadId) return null
+      const row = await prisma.generationRun.findFirst({ where: scope, orderBy: [{ startedAt: "desc" }, { id: "desc" }], select: runSelect })
+      return row ? mapRow(row) : null
+    },
+  })
+}
+
+const MAX_IMAGE_BYTES = 20 * 1024 * 1024
+async function imageDataUrl(source: string): Promise<string> {
+  if (source.startsWith("data:")) {
+    if (!/^data:image\/(png|jpeg|webp);base64,[A-Za-z0-9+/]+={0,2}$/.test(source)) throw new Error("Invalid image data")
+    if (Buffer.byteLength(source.split(",")[1], "base64") > MAX_IMAGE_BYTES) throw new Error("Image too large")
+    return source
+  }
+  const url = assertFetchableUrl(source)
+  const response = await fetch(url, { redirect: "error", signal: AbortSignal.timeout(30_000) })
+  if (!response.ok || !response.body) throw new Error("Image download failed")
+  const mime = response.headers.get("content-type")?.split(";")[0]
+  if (!mime || !["image/png", "image/jpeg", "image/webp"].includes(mime)) throw new Error("Invalid image content type")
+  const reader = response.body.getReader()
+  const chunks: Uint8Array[] = []
+  let size = 0
+  try {
+    while (true) {
+      const next = await reader.read()
+      if (next.done) break
+      size += next.value.byteLength
+      if (size > MAX_IMAGE_BYTES) throw new Error("Image too large")
+      chunks.push(next.value)
+    }
+  } finally {
+    await reader.cancel()
+  }
+  if (!size) throw new Error("Empty image")
+  return `data:${mime};base64,${Buffer.concat(chunks).toString("base64")}`
+}
+
+export async function generateImageForUser(opts: MediaGenOptions & {
+  prompt: string; numberOfImages?: number; size?: string; referenceImage?: string; sourceRunId?: string
+}) {
+  await requireThread(opts)
+  const adapter = resolveImageAdapter(opts)
+  if (adapter.kind === "unavailable") throw new Error(`Image generation is not available for adapter kind: ${opts.adapterKind}`)
+  const reference = opts.referenceImage ? await imageDataUrl(opts.referenceImage) : undefined
+  const runId = randomUUID()
+  const store = createGenerationRunStore(opts, { prompt: opts.prompt, sourceRunId: opts.sourceRunId, numberOfImages: opts.numberOfImages, size: opts.size })
+  const result = await generateImage({
+    adapter, runId, threadId: opts.threadId,
+    prompt: reference ? [{ type: "text", content: opts.prompt }, { type: "image", source: { type: "url", value: reference } }] : opts.prompt,
+    ...(opts.numberOfImages ? { numberOfImages: opts.numberOfImages } : {}),
+    ...(opts.size ? { size: opts.size } : {}),
+    middleware: [{
+      name: "image-result-storage",
+      onStart(ctx) {
+        ctx.resultTransforms?.push(async raw => {
+          const result = raw as ImageGenerationResult
+          if (!result.images?.length) throw new Error("No images generated")
+          const artifacts: PersistedArtifactRef[] = []
+          for (const [index, image] of result.images.entries()) {
+            const source = image.b64Json ? `data:image/png;base64,${image.b64Json}` : image.url
+            if (!source) throw new Error("Generated image is unavailable")
+            const url = await imageDataUrl(source)
+            const mimeType = url.slice(5, url.indexOf(";"))
+            artifacts.push({ role: "output", artifactId: `${runId}-${index}`, threadId: opts.threadId, runId,
+              name: `image-${index}`, mimeType, size: Buffer.byteLength(url.split(",")[1], "base64"),
+              createdAt: new Date().toISOString(), url,
+              source: { activity: "image", path: `images[${index}]`, provider: adapter.name, model: adapter.model, mediaType: "image" },
+            })
+          }
+          return { ...result, artifacts }
+        })
+      },
+    }, withGenerationPersistence({ stores: { generationRuns: store } })],
+  })
+  return { ...result, runId }
+}
+
+export async function generateVideoForUser(opts: MediaGenOptions & { prompt: string; duration?: number; size?: string }) {
+  await requireThread(opts)
+  const adapter = resolveVideoAdapter(opts)
+  if (adapter.kind === "unavailable") throw new Error(`Video generation is not available for adapter kind: ${opts.adapterKind}`)
+  const result = await generateVideo({ adapter, prompt: opts.prompt, duration: opts.duration, size: opts.size, threadId: opts.threadId,
+    middleware: [withGenerationPersistence({ stores: { generationRuns: createGenerationRunStore(opts, { prompt: opts.prompt }) } })],
+  })
+  return { ...result, runId: videoRunStoreKey(adapter.name, result.jobId) }
+}
+
+export async function pollVideoForUser(opts: MediaGenOptions & { jobId: string }) {
+  await requireThread(opts)
+  const adapter = resolveVideoAdapter(opts)
+  if (adapter.kind === "unavailable") throw new Error(`Video generation is not available for adapter kind: ${opts.adapterKind}`)
+  const store = createGenerationRunStore(opts)
+  const run = await store.get(videoRunStoreKey(adapter.name, opts.jobId))
+  if (!run || run.activity !== "video" || run.model !== opts.model) throw new Error("Video run not found")
+  return getVideoJobStatus({ adapter, jobId: opts.jobId, threadId: opts.threadId,
+    middleware: [withGenerationPersistence({ stores: { generationRuns: store } })],
+  })
+}
+
+export async function generateSpeechForUser(opts: MediaGenOptions & { text: string; voice?: string; format?: "mp3" | "opus" | "aac" | "flac" | "wav" | "pcm" }) {
+  await requireThread(opts)
+  const adapter = resolveSpeechAdapter(opts)
+  if (adapter.kind === "unavailable") throw new Error(`Speech generation is not available for adapter kind: ${opts.adapterKind}`)
+  const runId = randomUUID()
+  const result = await generateSpeech({ adapter, runId, text: opts.text, voice: opts.voice, format: opts.format, threadId: opts.threadId,
+    middleware: [withGenerationPersistence({ stores: { generationRuns: createGenerationRunStore(opts, { text: opts.text }) } })],
+  })
+  return { ...result, runId }
+}
+
+export async function generateTranscriptionForUser(opts: MediaGenOptions & { audio: string | File | Blob | ArrayBuffer; language?: string; prompt?: string }) {
+  await requireThread(opts)
+  const adapter = resolveTranscriptionAdapter(opts)
+  if (adapter.kind === "unavailable") throw new Error(`Transcription is not available for adapter kind: ${opts.adapterKind}`)
+  return generateTranscription({ adapter, audio: opts.audio, language: opts.language, prompt: opts.prompt, threadId: opts.threadId,
+    middleware: [withGenerationPersistence({ stores: { generationRuns: createGenerationRunStore(opts) } })],
+  })
+}
+
+export interface GeneratePrototypeInput {
+  userId: string
+  threadId: string
+  prompt: string
+  model?: string
+}
+
+/** spec 路线生成（P4e §3.1，ADR 39）：复用 prd.prototype bizKey 链，组件树落 GenerationRun。
+ *  落库走 recordGenerationRun（runChatAsAgent 无 activity middleware 挂点）；
+ *  token 计量走 runChat 既有 AiUsageLog 链，不占 mediaGenerationsMonthly。 */
+export async function generatePrototypeForUser(opts: GeneratePrototypeInput) {
+  await requireThread(opts)
+  const raw = await runChatAsAgent("prd.prototype", [{ role: "user", content: opts.prompt }], {
+    userId: opts.userId, model: opts.model, promptVars: { catalog: "（可用组件库为空，全部使用 custom 生成）" },
+    outputSchema: prototypeSchema as unknown as SchemaInput,
+  })
+  let result: GeneratePrototypeResult
+  try {
+    result = prototypeSchema.parse(JSON.parse(stripCodeFence(raw))) as unknown as GeneratePrototypeResult
+  } catch {
+    throw new Error("AI 返回内容无法解析为 JSON，请重试")
+  }
+  const run = await recordGenerationRun({
+    userId: opts.userId, threadId: opts.threadId, activity: "prototype", status: "completed",
+    result, provider: "agent", model: opts.model ?? "agent-default",
+    usage: {}, artifacts: [],
+  })
+  return { runId: run.id, result, artifacts: [] }
+}

package/src/server/ai-gateway/media/prototype-schema.ts

@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: AGPL-3.0-or-later
+// AllBlue - 理想之海
+// Copyright (C) 2026 AllBlue Contributors
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <https://www.gnu.org/licenses/>.
+
+import { z } from "zod"
+
+export type JsonValue =
+  | string
+  | number
+  | boolean
+  | null
+  | JsonValue[]
+  | { [key: string]: JsonValue }
+
+const jsonValueSchema: z.ZodType<JsonValue> = z.lazy(() =>
+  z.union([
+    z.string(),
+    z.number(),
+    z.boolean(),
+    z.null(),
+    z.array(jsonValueSchema),
+    z.record(z.string(), jsonValueSchema),
+  ]),
+)
+
+export const prototypeSchema = z.object({
+  message: z.string().describe("生成结果的说明文字"),
+  components: z.array(
+    z.object({
+      type: z.string().min(1).describe("组件类型标识（对应 catalog 的 type 字段或 'custom'）"),
+      source: z.enum(["builtin", "custom"]).describe("来源：builtin=组件库 / custom=全新生成"),
+      name: z.string().min(1).describe("组件展示名称"),
+      props: z.record(z.string(), jsonValueSchema).optional(),
+      width: z.number().optional(),
+      height: z.number().optional(),
+    }),
+  ).describe("组件列表"),
+})
+
+/** AI 生成的单个组件规格 */
+export interface GeneratedComponent {
+  type: string
+  source: "builtin" | "custom"
+  name: string
+  props?: Record<string, JsonValue>
+  width?: number
+  height?: number
+}
+
+export interface GeneratePrototypeResult {
+  message: string
+  components: GeneratedComponent[]
+}
+
+/** 清理 LLM 输出中可能残留的 markdown 代码块标记 */
+export function stripCodeFence(raw: string): string {
+  let s = raw.trim()
+  if (s.startsWith("```")) {
+    s = s.replace(/^```(?:json)?\s*\n?/, "").replace(/\n?```\s*$/, "")
+  }
+  return s.trim()
+}

package/src/server/ai-gateway/media/recorder.ts

@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: AGPL-3.0-or-later
+// AllBlue - 理想之海
+// Copyright (C) 2026 AllBlue Contributors
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <https://www.gnu.org/licenses/>.
+
+import { prisma, type Prisma } from "@allbluecn/database"
+
+/**
+ * P3 遗留落库函数，被 GenerationRunStore（generation-persistence.ts）内部调用。
+ *
+ * 保留作为兜底：直接写入 generation_runs 表，不经过 withGenerationPersistence。
+ * 业务代码应优先使用 store 接口。
+ */
+
+export type GenerationRunStatus = "running" | "completed" | "failed" | "interrupted"
+
+export interface RecordGenerationRunInput {
+  threadId: string
+  userId: string
+  /** 官方 GenerationRunRecord 视角：chat|image|audio|tts|video|transcription|sandbox */
+  activity: string
+  provider?: string
+  model?: string
+  status: GenerationRunStatus
+  result?: unknown
+  artifacts?: unknown
+  usage?: unknown
+  error?: string
+}
+
+function toJson(value: unknown): Prisma.InputJsonValue {
+  return value as unknown as Prisma.InputJsonValue
+}
+
+/** 终态写入 finishedAt（与 startedAt 同值，二者成对才有时长可算）。 */
+const TERMINAL_STATUSES: ReadonlySet<GenerationRunStatus> = new Set([
+  "completed",
+  "failed",
+  "interrupted",
+])
+
+export async function recordGenerationRun(input: RecordGenerationRunInput) {
+  const now = new Date()
+  return prisma.generationRun.create({
+    data: {
+      threadId: input.threadId,
+      userId: input.userId,
+      activity: input.activity,
+      kind: input.activity,
+      status: input.status,
+      provider: input.provider,
+      model: input.model,
+      result: input.result === undefined ? undefined : toJson(input.result),
+      artifacts: input.artifacts === undefined ? undefined : toJson(input.artifacts),
+      usage: input.usage === undefined ? undefined : toJson(input.usage),
+      error: input.error,
+      startedAt: now,
+      finishedAt: TERMINAL_STATUSES.has(input.status) ? now : null,
+    },
+  })
+}

package/src/server/ai-gateway/memory/index.ts

@@ -0,0 +1,76 @@
+// SPDX-License-Identifier: AGPL-3.0-or-later
+// AllBlue - 理想之海
+// Copyright (C) 2026 AllBlue Contributors
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Affero General Public License for more details.
+//
+// You should have received a copy of the GNU Affero General Public License
+// along with this program. If not, see <https://www.gnu.org/licenses/>.
+
+import type { ChatMiddleware } from "@tanstack/ai"
+import { memoryMiddleware, type MemoryMiddlewareRole } from "@tanstack/ai-memory"
+import {
+  createPgVectorMemoryAdapter,
+  type PgVectorMemoryAdapterOptions,
+} from "./pgvector-adapter"
+import { createMemoryScopeResolver } from "./scope"
+
+const TELEMETRY_ERROR_MAX_CHARS = 200
+
+function truncateError(error: string | undefined): string {
+  if (!error) return ""
+  return error.length > TELEMETRY_ERROR_MAX_CHARS
+    ? `${error.slice(0, TELEMETRY_ERROR_MAX_CHARS)}...`
+    : error
+}
+
+// 桌面端 SQLite 无 pgvector，直接不挂（与 unified-router 的 desktop 短路一致）
+export function isMemoryMiddlewareEnabled(): boolean {
+  if (process.env.DEPLOY_TARGET === "desktop") return false
+  if (process.env.DATABASE_PROVIDER === "sqlite") return false
+  return true
+}
+
+// 默认 recall+save：recall 侧已完整实现，且 pgvector adapter 有零记忆快路径
+// （该用户尚无沉淀时不发 embedding 调用），save-only 只作退化开关。
+export function resolveMemoryRole(): MemoryMiddlewareRole {
+  return process.env.AI_MEMORY_ROLE === "save-only" ? "save-only" : "recall+save"
+}
+
+export function buildMemoryMiddleware(
+  userId: string,
+  adapterOptions: PgVectorMemoryAdapterOptions = {},
+): ChatMiddleware | null {
+  if (!isMemoryMiddlewareEnabled()) return null
+
+  return memoryMiddleware({
+    adapter: createPgVectorMemoryAdapter(adapterOptions),
+    scope: createMemoryScopeResolver(userId),
+    role: resolveMemoryRole(),
+    onRecall: async ({ scope, query, result }) => {
+      console.log(
+        `[AI Memory] recall userId=${scope.userId} threadId=${scope.threadId} fragments=${result.fragments?.length ?? 0} promptChars=${result.systemPrompt.length} queryChars=${query.length}`,
+      )
+    },
+    onSave: async ({ scope, receipts }) => {
+      const okCount = receipts.filter((receipt) => receipt.ok).length
+      console.log(
+        `[AI Memory] save userId=${scope.userId} threadId=${scope.threadId} wrote=${okCount}/${receipts.length}`,
+      )
+      for (const receipt of receipts) {
+        if (receipt.ok) continue
+        console.warn(
+          `[AI Memory] save failed userId=${scope.userId}: ${truncateError(receipt.error)}`,
+        )
+      }
+    },
+  })
+}