Technical Requirements for the Universal Data Bot (Short Version) 1. The bot acts as a full employee The bot must: • Read and return data strictly from connected tables • Add, update, and delete rows/cells • Create new sheets (tabs) when requested The bot must not invent data — only use what exists in the tables. 2. User requests are free-form The user can speak naturally. Bot must interpret meaning, not wait for strict syntax. Rules: • 1 exact match → answer immediately • Multiple matches → show buttons for selection • No matches → politely say no data 3. When buttons appear — use button text only After a button is clicked, the bot must work strictly with the button value, not the original user query. 4. Multi-level intelligent search The bot processes requests step-by-step: Choose the correct table Choose the correct sheet/tab Find the correct row (names/materials/items etc.) Find the correct column (dates, totals, amounts, hours…) Return the correct cell value with clean formatting If something is unclear — the bot must ask for clarification. 5. Context awareness The bot must understand conversation context across all tables. After every successful GET_VALUE: → Save LAST_FOUND (table, sheet, full_row_key, column). During UPDATE_DATA, if user doesn’t specify a name — the bot must use saved context. Context must be injected BEFORE sending prompt to the AI and applied AFTER intent detection. Bot must: ✓ Work with any type of data (names, materials, objects, inventory) ✓ Keep context across any table ✓ Always reuse exact saved values ✓ Treat this rule as critical 6. Behavioral rules The bot must not: • Invent data • Switch to another sheet unless user chooses it • Skip clarification steps The bot must: • Understand natural phrases (“on the 11th”, “for July”, “how much he worked yesterday”) • Ask when a request is ambiguous • Operate only inside the active sheet • Behave like a careful real employee