File sandbox
File profiling runs in an isolated container, without networking and with the main filesystem separated.
Secure AI
Business spreadsheets are not just any prompt.
Excel and CSV files contain customers, prices, defects, complaints, costs, suppliers and operational decisions. TalkToMyExcel puts a technical boundary around AI before generating an answer.
Boundary
Four controls before the answer.
Separate workspaces
Each user works on dedicated datasets, DuckDB storage and semantic index.
Controlled provider
Regolo.ai is the recommended choice for European inference, zero retention and no training on customer data.
Reduced evidence
The model receives only the context needed for the answer, not the whole file without criteria.
Architecture
The model is not the database.
DuckDB handles structured queries, filters and counts. Chroma handles text similarity on selected columns. The LLM enters at the end, once compact evidence already exists.
This reduces hallucination risk and makes it easier to understand where an answer comes from: rows, tables and columns remain part of the conversation.
Regolo.ai recommended
European sovereignty and zero retention for important data.
With Regolo.ai, prompts and outputs are handled in memory by the inference layer and are not persisted. Customer request content is not used to train or fine-tune shared models.
Details on Regolo.aiWhat to avoid
The difference from a generic chatbot.
Generic chatbot
- File copied into the prompt without structure.
- Provider chosen only for convenience or cost.
- Little visibility on retention and training.
- Answers difficult to trace back to rows.
TalkToMyExcel
- File profiled and imported as a dataset.
- Regolo.ai recommended for compliance and zero retention.
- Hybrid engine with selected evidence.
- Cited answers that teams can check.
Assessment
Bring the most delicate use case.
If it works on sensitive data, it also works on simple data.