feat(server): improve context metadata & matching (#12064)

fix AI-20

<!-- This is an auto-generated comment: release notes by coderabbit.ai -->
## Summary by CodeRabbit

- **New Features**
  - Enhanced file metadata with MIME type, blob ID, and file name across context and workspace, now visible in UI and API.
  - Added workspace-level matching for files and documents with configurable thresholds and workspace scoping in search queries.
  - Introduced a new error type and user-friendly messaging for global workspace context matching failures.

- **Bug Fixes**
  - Improved consistent handling of file MIME types and nullable context IDs for accurate metadata.

- **Documentation**
  - Updated GraphQL schema, queries, and mutations to include new metadata fields, optional parameters, and error types.

- **Style**
  - Added new localization strings for global context matching error messages.

- **Tests**
  - Extended test coverage with new and updated snapshot tests for metadata and matching logic.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

packages/backend/server/src/models/common/copilot.ts

@@ -46,6 +46,7 @@ export const ContextFileSchema = z.object({
   id: z.string(),
   chunkSize: z.number(),
   name: z.string(),
+  mimeType: z.string().optional(),
   status: ContextEmbedStatusSchema,
   error: z.string().nullable(),
   blobId: z.string(),
@@ -100,6 +101,9 @@ export type ChunkSimilarity = {
 
 export type FileChunkSimilarity = ChunkSimilarity & {
   fileId: string;
+  blobId: string;
+  name: string;
+  mimeType: string;
 };
 
 export type DocChunkSimilarity = ChunkSimilarity & {
@@ -108,6 +112,7 @@ export type DocChunkSimilarity = ChunkSimilarity & {
 
 export const CopilotWorkspaceFileSchema = z.object({
   fileName: z.string(),
+  blobId: z.string(),
   mimeType: z.string(),
   size: z.number(),
 });

packages/backend/server/src/models/copilot-context.ts

@@ -1,6 +1,7 @@
 import { randomUUID } from 'node:crypto';
 
 import { Injectable } from '@nestjs/common';
+import { Transactional } from '@nestjs-cls/transactional';
 import { Prisma } from '@prisma/client';
 
 import { CopilotSessionNotFound } from '../base';
@@ -179,9 +180,9 @@ export class CopilotContextModel extends BaseModel {
     contextId: string,
     topK: number,
     threshold: number
-  ): Promise<FileChunkSimilarity[]> {
+  ): Promise<Omit<FileChunkSimilarity, 'blobId' | 'name' | 'mimeType'>[]> {
     const similarityChunks = await this.db.$queryRaw<
-      Array<FileChunkSimilarity>
+      Array<Omit<FileChunkSimilarity, 'blobId' | 'name' | 'mimeType'>>
     >`
       SELECT "file_id" as "fileId", "chunk", "content", "embedding" <=> ${embedding}::vector as "distance" 
       FROM "ai_context_embeddings"
@@ -217,6 +218,7 @@ export class CopilotContextModel extends BaseModel {
     });
   }
 
+  @Transactional()
   async matchWorkspaceEmbedding(
     embedding: number[],
     workspaceId: string,
@@ -232,6 +234,18 @@ export class CopilotContextModel extends BaseModel {
       ORDER BY "distance" ASC
       LIMIT ${topK};
     `;
-    return similarityChunks.filter(c => Number(c.distance) <= threshold);
+
+    const matchedChunks = similarityChunks.filter(
+      c => Number(c.distance) <= threshold
+    );
+    const matchedDocIds = Array.from(new Set(matchedChunks.map(c => c.docId)));
+    if (!matchDocIds?.length && matchedDocIds.length) {
+      const ignoredDocs = await this.models.copilotWorkspace.checkIgnoredDocs(
+        workspaceId,
+        matchedDocIds
+      );
+      return matchedChunks.filter(c => !ignoredDocs.includes(c.docId));
+    }
+    return matchedChunks;
   }
 }

packages/backend/server/src/models/copilot-workspace.ts

@@ -16,6 +16,25 @@ import type {
 
 @Injectable()
 export class CopilotWorkspaceConfigModel extends BaseModel {
+  @Transactional()
+  private async listIgnoredDocIds(
+    workspaceId: string,
+    options?: PaginationInput
+  ) {
+    return await this.db.aiWorkspaceIgnoredDocs.findMany({
+      where: {
+        workspaceId,
+      },
+      select: {
+        docId: true,
+        createdAt: true,
+      },
+      orderBy: { createdAt: 'desc' },
+      skip: options?.offset,
+      take: options?.first,
+    });
+  }
+
   @Transactional()
   async updateIgnoredDocs(
     workspaceId: string,
@@ -23,7 +42,7 @@ export class CopilotWorkspaceConfigModel extends BaseModel {
     remove: string[] = []
   ) {
     const removed = new Set(remove);
-    const ignored = await this.listIgnoredDocs(workspaceId).then(
+    const ignored = await this.listIgnoredDocIds(workspaceId).then(
       r => new Set(r.map(r => r.docId).filter(id => !removed.has(id)))
     );
     const added = add.filter(id => !ignored.has(id));
@@ -51,25 +70,11 @@ export class CopilotWorkspaceConfigModel extends BaseModel {
     return added.length + ignored.size;
   }
 
-  @Transactional()
   async listIgnoredDocs(
     workspaceId: string,
-    options?: {
-      includeRead?: boolean;
-    } & PaginationInput
+    options?: PaginationInput
   ): Promise<IgnoredDoc[]> {
-    const row = await this.db.aiWorkspaceIgnoredDocs.findMany({
-      where: {
-        workspaceId,
-      },
-      select: {
-        docId: true,
-        createdAt: true,
-      },
-      orderBy: { createdAt: 'desc' },
-      skip: options?.offset,
-      take: options?.first,
-    });
+    const row = await this.listIgnoredDocIds(workspaceId, options);
     const ids = row.map(r => ({ workspaceId, docId: r.docId }));
     const docs = await this.models.doc.findMetas(ids);
     const docsMap = new Map(
@@ -106,7 +111,7 @@ export class CopilotWorkspaceConfigModel extends BaseModel {
 
   @Transactional()
   async checkIgnoredDocs(workspaceId: string, docIds: string[]) {
-    const ignored = await this.listIgnoredDocs(workspaceId).then(
+    const ignored = await this.listIgnoredDocIds(workspaceId).then(
       r => new Set(r.map(r => r.docId))
     );
 
@@ -214,9 +219,19 @@ export class CopilotWorkspaceConfigModel extends BaseModel {
     const similarityChunks = await this.db.$queryRaw<
       Array<FileChunkSimilarity>
     >`
-      SELECT "file_id" as "fileId", "chunk", "content", "embedding" <=> ${embedding}::vector as "distance" 
-      FROM "ai_workspace_file_embeddings"
-      WHERE workspace_id = ${workspaceId}
+      SELECT
+        e."file_id" as "fileId",
+        f."file_name" as "name",
+        f."blob_id" as "blobId",
+        f."mime_type" as "mimeType",
+        e."chunk",
+        e."content",
+        e."embedding" <=> ${embedding}::vector as "distance" 
+      FROM "ai_workspace_file_embeddings" e
+      JOIN "ai_workspace_files" f
+        ON e."workspace_id" = f."workspace_id"
+        AND e."file_id" = f."file_id"
+      WHERE e.workspace_id = ${workspaceId}
       ORDER BY "distance" ASC
       LIMIT ${topK};
     `;