​
Overview

An Excel agent combines Chunkr’s parsing power with an interactive viewer and LLM capabilities. Users can view native Excel cells, search for tables, and let the LLM read/write to specific cell ranges while keeping everything synchronized.

​
Architecture

​
How It Works

​
1. Chunkr Output as Foundation

  • Chunkr parses the Excel file once into structured data
  • The viewer renders native cells directly from this parsed output
  • All components work with the same underlying data structure

​
2. LLM Integration

  • Read Operations: LLM analyzes specific cell ranges for patterns and insights
  • Write Operations: LLM requests data changes through the sync code
  • No Re-parsing: Changes are made to existing parsed data, Chunkr doesn’t re-parse

​
3. Sync Code Coordination

  • Sync code is the central coordinator for all data modifications
  • LLM sends change requests to sync code rather than modifying data directly
  • Sync code updates parsed data, viewer, and vector DB embeddings simultaneously
  • Ensures all components stay synchronized with a single source of truth

​
4. Search & Discovery

  • Chunkr parsed output gets converted into embeddings for semantic understanding
  • Vector DB stores these embeddings for efficient similarity search
  • Enables semantic search for more intelligent data discovery
  • LLM can quickly locate specific data ranges through natural language queries

​
Example Workflow

  1. Upload Excel file → Chunkr parses structure
  2. Create embeddings → Convert parsed data to vectors for search
  3. View native cells → Viewer displays from parsed output
  4. Search for tables → Find specific tables through natural language in vector DB
  5. LLM operations → Read/write to cell ranges
  6. Auto-update → Viewer reflects changes immediately
The agent acts as the central coordinator, ensuring seamless interaction between the user interface and the underlying tools while maintaining data consistency through the Chunkr output.