Spreadsheet Governance & Collaborative AI Workflows
In-App 2 MB Attachment vs MCP 200-Cell Read Audit
Choose the right inspection path in Arte Sheets: metered in-app Pro Assistant file excerpts (.xlsx/.csv/.tsv/.txt <= 2 MB, safe non-executing) versus external MCP read_range (<= 200 cells raw values/formulas + revision, 0 Arte AI credits).
Understand exact size caps, formula evaluation behavior, and credit rules across both AI read paths!
Teams inspecting financial schedules in Arte Sheets have two distinct, strictly bounded AI context paths with different capabilities and billing semantics! Inside the browser editor, the **In-App Pro Assistant** accepts `.xlsx`, `.csv`, `.tsv`, and `.txt` file attachments up to **2 MB** (`2,097,152` bytes) parsed strictly as safe, non-executing text/cell excerpts (never auto-running macros or overwriting sheets without preview/Undo or Suggest-mode routing), requiring an active Pro plan and consuming metered monthly **Arte AI credits** (`2,000` monthly grant) when the request is sent. By contrast, an external scoped OAuth client (such as the Arte Sheets ChatGPT `0.1.0` connector — currently `In review` / `Not published`; revocable anytime under `Profile -> Connected apps`) inspects a **cloud collaboration workbook** (local browser-only files are unavailable) via `read_range` (`sheets:read`), capped at **`<= 200` cells** (`rows * cols <= 200`) per call, returning sheet IDs, current `revision`, whole-cell comment anchors, and **raw stored inputs/formulas WITHOUT recalculating formula results**, while consuming **`0` Arte AI credits** under the shared `120` MCP tool calls/min per-account limiter!
Build an 8-row preflight calculator to verify chunking and attachment gates! Suppose an analyst wants to inspect a `15` row x `12` column cloud schedule (`180` cells <= `200`, B2 & B3) alongside an external `.csv` vendor extract of `1,572,864` bytes (`1.50` MB <= `2.00` MB = `2,097,152` bytes, B4 with Supported_Ext_1_0 = `1` in B5 and Cloud_Workbook_1_0 = `1` in B6): `read_range` Call Count needed for `15 x 12 = 180` cells is `CEILING_MATH(180 / 200) = 1` call (`180 <= 200` -> single call, `0` Arte AI credits!), whereas a `35` row x `10` column schedule (`350` cells) requires `CEILING_MATH(350 / 200) = 2` paginated `read_range` calls! Meanwhile, the `1.50` MB `.csv` passes the In-App Attachment Gate (`524,288` bytes = `0.50` MB headroom below `2,097,152`)!
| Inspection Governance Dimension | In-App Pro Assistant Attachment | MCP Connector read_range (0.1.0 In Review) |
|---|---|---|
| Supported Source & Payload Cap | .xlsx/.csv/.tsv/.txt <= 2,097,152 bytes | Cloud Workbook Range <= 200 cells / call |
| Formula & Execution Semantics | Safe non-executing excerpt + live sheet context | Raw inputs/formulas + revision (No Recalc!) |
| Arte AI Credit Consumption | Metered monthly Pro AI credits per send | 0 Arte AI credits (120 calls/min burst cap) |
| 15x12 Range (180 cells) & 1.5 MB CSV | Attachment Valid (0.50 MB Headroom) | Single read_range Call (20 cells headroom) |
MCP read_range Required Paginated Call Count Formula (<= 200 Cells / Call)
Validate Cloud_Workbook = 1 and positive range dimensions R >= 1, C >= 1 before computing CEILING_MATH((R * C) / 200).
=LET(r_ct, B2, c_ct, B3, cl_wb, B6, IF(OR(r_ct < 1, INT(r_ct) <> r_ct, c_ct < 1, INT(c_ct) <> c_ct), "Invalid Range Dimensions", IF(cl_wb <> 1, "Unavailable: Local Browser-Only Workbook", CEILING_MATH((r_ct * c_ct) / 200))))In-App Pro Assistant 2 MB Safe Attachment Gate Formula
Enforce both allowed file format (.xlsx/.csv/.tsv/.txt -> B5 = 1) and strict 2,097,152-byte ceiling.
=LET(by_sz, B4, ex_ok, B5, IF(ex_ok <> 1, "Rejected: Require .xlsx/.csv/.tsv/.txt", IF(OR(by_sz <= 0, by_sz > 2097152), "Rejected: Exceeds 2 MB Attachment Cap", "Valid Safe Non-Executing Excerpt")))Try the example
Use the original values or change them to test your understanding.