feat: add configurable file metadata exposure to LLMs

[acancino-gnv]

Summary

• Adds configurable file metadata exposure to LLMs when files are uploaded
• Metadata (filename, mimeType, sizeBytes, etc.) is injected into messages sent to LLMs
• Enables models to reference files by name and pass metadata to tools/MCP servers

Changes

• Added FileMetadataFields enum with core and opt-in fields
• Added fileMetadataConfigSchema for YAML configuration
• Implemented formatFileMetadata() function with markdown/json/xml output formats
• Updated extractFileContext() to inject metadata into messages
• Added comprehensive test coverage

Configuration Example

fileConfig:
  metadata:
    enabled: true
    fields:
      - filename
      - type
      - bytes
      - source
      - filepath  # opt-in: exposes storage path
    format: markdown  # or json, xml

Output Formats

Markdown (default):

**File Metadata:**
- **filename**: report.pdf
- **type**: application/pdf
- **bytes**: 1048576
- **size_human**: 1 MB

JSON:

{
  "filename": "report.pdf",
  "type": "application/pdf",
  "bytes": 1048576,
  "size_human": "1 MB"
}

Available Fields

Field	Description	Default
filename	Original filename	✅
type	MIME type	✅
bytes	Size in bytes (+ human readable)	✅
source	Storage backend (local, s3, azure, etc.)	❌
width / height	Image dimensions	❌
createdAt / updatedAt	Timestamps	❌
filepath	Full storage path (opt-in)	❌
conversationId	Session identifier (opt-in)	❌
file_id	Unique file ID (opt-in)	❌

Test plan

• Unit tests for formatFileMetadata() covering all formats and fields
• Unit tests for extractFileContext() with metadata enabled/disabled
• Manual testing with actual file uploads
• Verify metadata appears in LLM context for OpenAI, Anthropic, Google endpoints

[Copilot]

Pull request overview

This PR adds a configurable file metadata exposure feature that allows LLMs to access file metadata (filename, type, size, etc.) when processing attachments. Metadata can be injected in markdown, JSON, or XML formats, with configurable field selection including both safe defaults and opt-in sensitive fields.

Key changes:

• Adds type definitions and Zod schemas for file metadata configuration
• Implements metadata formatting functions supporting markdown, JSON, and XML output
• Integrates metadata injection into the file context extraction pipeline
• Provides comprehensive test coverage for formatting and extraction logic

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
packages/data-provider/src/types/files.ts	Adds FileMetadataConfig type definition for metadata configuration
packages/data-provider/src/file-config.ts	Defines FileMetadataFields enum, default fields, Zod schema, and merging logic
packages/api/src/files/context.ts	Implements metadata formatting functions and integrates metadata injection into extractFileContext
packages/api/src/files/context.spec.ts	Adds comprehensive unit tests for metadata formatting and file context extraction
librechat.example.yaml	Documents configuration options with examples of safe and opt-in fields

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[alonso-cancino]

@danny-avila I worked out the copilot comments, if you could review it again.

[acancino-gnv]

@danny-avila can you review it again please, I fixed all copilot issues, and made manual tests on my end to see the metadata injection works properly with different providers.