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 -->
This commit is contained in:
darkskygit
2025-05-14 06:32:29 +00:00
parent 04c5fd6dfc
commit cecf545590
36 changed files with 465 additions and 108 deletions

View File

@@ -148,6 +148,48 @@ export class CopilotContextService implements OnApplicationBootstrap {
return null;
}
async matchWorkspaceFiles(
workspaceId: string,
content: string,
topK: number = 5,
signal?: AbortSignal,
threshold: number = 0.5
) {
if (!this.embeddingClient) return [];
const embedding = await this.embeddingClient.getEmbedding(content, signal);
if (!embedding) return [];
const chunks = await this.models.copilotWorkspace.matchFileEmbedding(
workspaceId,
embedding,
topK * 2,
threshold
);
return this.embeddingClient.reRank(content, chunks, topK, signal);
}
async matchWorkspaceDocs(
workspaceId: string,
content: string,
topK: number = 5,
signal?: AbortSignal,
threshold: number = 0.5
) {
if (!this.embeddingClient) return [];
const embedding = await this.embeddingClient.getEmbedding(content, signal);
if (!embedding) return [];
const workspace = await this.models.copilotContext.matchWorkspaceEmbedding(
embedding,
workspaceId,
topK * 2,
threshold
);
return this.embeddingClient.reRank(content, workspace, topK);
}
@OnEvent('workspace.doc.embed.failed')
async onDocEmbedFailed({
contextId,