GPTSheet Documentation

BYOK LLM toolkit for Google Sheets and Excel. Pay once. Bring your own API key. Your data and keys never touch our servers.

Welcome

GPTSheet adds AI to your spreadsheets through three surfaces: cell formulas like =AI and =AI_CLASSIFY, a sidebar with batch tools and an API fetcher, and a chat agent (Premium) that can read and edit your sheet conversationally.

There are two plans:

Install

Google Sheets

1. In Google Sheets, open Extensions → Add-ons → Get add-ons and search for GPTSheet.
2. Click Install and pick the Google account you want to use.
3. Open any sheet → Extensions → GPTSheet → Open sidebar.

Microsoft Excel

1. In Excel, open Insert → Get Add-ins and search for GPTSheet.
2. Click Add and sign in with the Microsoft account you want to use.
3. Open Excel → Home → My Add-ins → GPTSheet.

One license covers both platforms.

Activate your license

After purchase you'll get an email with your license key, formatted like GS-XXXX-XXXX-XXXX-XXXX.

1. Open the sidebar → click the ≡ menu (top-left).
2. Tap License.
3. Paste the key and click Activate.

Set up API keys

GPTSheet is BYOK — you bring your own keys from each LLM provider you want to use. Keys are stored only in your browser's localStorage for sidebar tools, and in your sheet's PropertiesService for cell formulas. Our server never sees them.

1. Open the sidebar → ≡ menu → API keys.
2. Click the ✎ pencil next to a provider (OpenAI, Anthropic, Gemini).
3. Paste your key and click Save.

The first provider you save becomes the default — used by sidebar tools and as the default for cell formulas. You can change the default anytime by clicking Make default on another configured provider.

See the Providers section below for where to get each key.

Custom Prompt

The escape hatch for "do X to every row in this column" when no named formula fits.

1. Select a range in the sheet (e.g. A2:A20).
2. Sidebar → Prompt tab.
3. Type an instruction. Click Run on selection.

The sidebar reads your selection's first column, calls the LLM once per row, and writes results into the column immediately to the right.

Example: select A2:A20 filled with company names → type "Find each company's industry, one word" → results land in B2:B20.

Bulk Apply

Run a named formula across hundreds or thousands of rows with progress + stop.

1. Sidebar → Bulk tab.
2. Pick a formula from the dropdown (=AI_CLASSIFY, =AI_TRANSLATE, etc.).
3. Fill in the formula-specific parameters (categories, target language, etc.).
4. Set the input range and output column (auto-populated when you click Use selection).
5. Click Run. Watch the progress bar. Click Stop anytime — partial results are preserved.

API Fetcher

Paste a REST URL → JSON data appears in your sheet. Auth tokens stay in your browser; GPTSheet never sees them.

1. Click into an empty area of your sheet (the data will land at the active cell).
2. Sidebar → Fetch tab.
3. Paste the URL.
4. Optionally paste JSON headers (e.g. {"Authorization": "Bearer ..."}).
5. Click Fetch to sheet.

The fetcher handles common JSON shapes: array-of-objects (becomes a table with headers), array-of-primitives (single column), or a single object (two columns: key, value).

Cell formulas overview

Each formula calls an LLM from Google Apps Script's UrlFetchApp using your saved provider key. You can use them like any other Sheets formula — composing with ARRAYFORMULA, fill-down, conditional refs, etc.

Common parameters

Every formula accepts these optional trailing arguments:

=AI Basic

Generic LLM prompt that returns a single value.

=AI("Capital of France")          → Paris
=AI("Capital of " & A2)            → Tokyo (when A2 is "Japan")

=AI_LIST / =AI_HLIST Basic

Returns a list — one item per cell. AI_LIST fills vertically; AI_HLIST fills horizontally.

=AI_LIST("5 marketing channels for SaaS")

=AI_TABLE Basic

Returns a 2D table with headers in the first row.

=AI_TABLE("Top 3 EV makers", "brand,model,price")

=AI_CLASSIFY Basic

Picks one category from a list. Deterministic, taggable.

=AI_CLASSIFY(A2, "positive, negative, neutral")
=AI_CLASSIFY(A2, "enterprise, smb, mid-market", "Use company size as a hint")

=AI_EXTRACT Basic

Pulls specific elements out of messy text. Returns comma-separated values.

=AI_EXTRACT(A2, "phone numbers")
=AI_EXTRACT(A2, "email addresses")
=AI_EXTRACT(A2, "dates in ISO format")

=AI_TRANSLATE Basic

Translates text. Returns only the translation.

=AI_TRANSLATE(A2, "Japanese")
=AI_TRANSLATE(A2, "Japanese", , "Keep brand names in English")

=AI_LANG Basic

Detect the language of a text. Returns the language name in English.

=AI_LANG("Bonjour le monde")          → French
=AI_LANG(A2)                          → e.g. Japanese, Spanish, Arabic

=AI_EDIT Basic

Edits text — fix grammar, change tone, rewrite for an audience.

=AI_EDIT(A2)                          → fixes grammar/spelling by default
=AI_EDIT(A2, "make it more concise")
=AI_EDIT(A2, "rewrite for a 5th grader")

=AI_FORMAT Basic

Standardize dates, addresses, names, currencies — without writing regex.

=AI_FORMAT(A2, "ISO date")
=AI_FORMAT(A2, "USD with 2 decimals")
=AI_FORMAT(A2, "First Last")

=AI_TAG Basic

Apply tags from your taxonomy, or auto-suggest tags.

=AI_TAG(A2, "urgent, bug, feature, docs")
=AI_TAG(A2, , , 3)                    → auto-suggest 3 tags

=AI_SPLIT / =AI_HSPLIT Basic

Semantic split — by sentences, paragraphs, sections, etc. Returns one piece per cell (vertical or horizontal).

=AI_SPLIT(A2, "sentences")
=AI_SPLIT(A2, "paragraphs")

=AI_MATCH Basic

Semantic match between two ranges — VLOOKUP that understands meaning. Uses OpenAI embeddings under the hood.

=AI_MATCH(A2:A100, B2:B500)
=AI_MATCH(A2:A100, B2:B500, 0.7, TRUE)   → with confidence score in adjacent column

=AI_CREATE_PROMPT Basic

Helper that concatenates cells and ranges into a single prompt string. Useful for building prompts from sheet data.

=AI(AI_CREATE_PROMPT("Topic:", A2, "Audience:", B2, "Write a tweet."))

=AI_VISION Premium

Analyze an image given its URL — caption, classify, extract text, etc.

=AI_VISION("Describe in one sentence", A2)
=AI_VISION("Extract any text", A2, TRUE)

Uses OpenAI's gpt-4o-mini by default (vision-capable). Pass the image URL in column A and the prompt as the first argument.

=AI_WEB Premium

Web-grounded answer with citations — uses OpenAI's Responses API with the web_search_preview tool.

=AI_WEB("AAPL closing price today")
=AI_WEB("Latest stable Node.js version. Just the number.")

=AI_IMAGE Premium

Generate an image and return a URL. Wrap with =IMAGE(...) to render in a cell.

=IMAGE(AI_IMAGE("a green apple on white background"))
=IMAGE(AI_IMAGE("photo of a sunset", "", "gemini-2.5-flash-image"))

Defaults to OpenAI dall-e-3. Pass gemini-2.5-flash-image or imagen-3.0-generate-002 as the third arg to route through Gemini instead (more permissive rate limits).

Agent overview Premium

A conversational agent that reads your sheet, plans, calls tools, and writes results. Multi-step, context-aware, and editable.

Open the sidebar → Agent tab. Your current cell selection is auto-attached as context — the agent sees what data you're working with before you even ask.

Example prompts:

• "For each row in A2:A20, find the company's industry and write it to column B"
• "Create a bar chart from this data showing revenue by product"
• "What stands out in this data?"
• "Generate an image of a logo for a fintech startup and put it at E5"

Agent tools

The agent has 7 tools it can call:

The agent picks the appropriate tool based on the request — you don't need to mention tool names. Use the model picker in the composer to switch between providers for that conversation.

Example tasks

Lead qualification

"For each company in A2:A50, look up the industry and employee count. Write industry to B, count to C. Then score each as 'enterprise', 'mid-market', or 'smb' in column D based on count."

Data viz from selection

"Compare these regions visually."   (with A1:B5 selected: Region, Revenue)

Formula composition

"Compute the running average of column B and put it in column C."   (uses insert_formula)

BYOK explained

BYOK = Bring Your Own Key. You pay LLM providers directly at their API price. GPTSheet is the software layer; the actual LLM bill goes to OpenAI, Anthropic, or Google.

For a typical user running ~1,000 cells through gpt-4o-mini per month, the LLM bill is about $1–3/month. There's no markup or middleman.

In neither case does the key (or your prompts/data) pass through GPTSheet's servers.

OpenAI

Get a key from platform.openai.com/api-keys.

Supported features (via OpenAI key):

• All chat formulas (=AI, =AI_CLASSIFY, etc.) — uses gpt-4o-mini by default
• =AI_VISION — uses gpt-4o-mini (vision-capable)
• =AI_WEB — uses the Responses API with web_search_preview tool
• =AI_IMAGE — uses dall-e-3 by default
• =AI_MATCH — uses text-embedding-3-small

Anthropic

Get a key from console.anthropic.com/settings/keys.

Supported features (via Anthropic key):

• All chat formulas — uses claude-haiku-4-5 by default
• Sidebar agent — full tool-use support via Anthropic's tool-use API

Not yet supported via Anthropic key alone: =AI_VISION (uses OpenAI), =AI_WEB, =AI_IMAGE, =AI_MATCH. These always use OpenAI keys.

Gemini

Get a key from aistudio.google.com/app/apikey. Free tier is generous (15 requests/min).

Supported features (via Gemini key):

• All chat formulas — uses gemini-2.0-flash by default
• Sidebar agent — full tool-use via Gemini's functionDeclarations
• =AI_IMAGE with gemini-2.5-flash-image or imagen-* models

Connectors — overview & setup model

Connectors let the Chat Agent (and direct Run-now buttons in the sidebar) pull data from your apps + databases into Sheets. Each one is configured once in Menu → Connectors; auth lives in your browser's localStorage and is never sent to our servers — only to the upstream provider (Stripe, Notion, GA4, etc.).

Connectors fall into three auth families:

• API token — paste a key from the provider's dashboard. Used by Stripe, Notion, Airtable, PostHog, Amplitude, Ghost.
• Google sign-in (one click) — used by GA4, Search Console, YouTube Insights. No JSON to paste, no service account to create — GPTSheet uses your existing Google identity (the one you're signed into Sheets with) via Apps Script's built-in OAuth.
• Direct database — host + credentials for direct SQL access. Used by MySQL, MS SQL Server, Postgres, DynamoDB. Supabase uses its PostgREST gateway + service_role key (no direct JDBC needed).

For the Google sign-in connectors, GPTSheet has no OAuth client of its own — Google's tokens are managed by Apps Script's runtime and never accessible to our code or our servers. This is structurally stronger than the service-account model we previously used.

Stripe

Connect Stripe to pull customers, subscriptions, charges, invoices, and balance transactions. The agent can answer "what's my MRR this month?" or "list churned customers last week."

Setup

1. Stripe Dashboard → Developers → API keys → Create restricted key.
2. Give it read-only permissions for Customers, Subscriptions, Charges, Invoices, Balance.
3. Copy the key (starts with rk_live_ or rk_test_).
4. In GPTSheet → Menu → Connectors → Stripe → paste the key → Connect.

Google Analytics (GA4)

Pull GA4 sessions, conversions, channels, top pages.

Setup

1. Make sure the Google account you're signed into Sheets with already has at least Viewer access to the GA4 property you want to query. (If not, ask your GA4 admin to add you under GA4 → Admin → Property Access Management → Add user.)
2. In GPTSheet → Menu → Connectors → Google Analytics → click Connect.
3. If this is the first Google connector you've used, Google will prompt you to authorize GPTSheet to view your Analytics data. Click Allow. (This is a one-time consent — it covers GA4, GSC, YouTube Insights, and Google Ads together.)
4. The Property dropdown auto-populates with every GA4 property your Google account can see. Pick one → Save changes.
5. Done. Run any GA4 action (List sessions, Conversion overview, Custom report, etc.).

Want to query multiple GA4 properties? Click "+ Add another account" and pick a different property. Each saved account binds to one property; you can have as many as you want.

Google Search Console

Top queries, top pages, country breakdowns.

Setup

1. Make sure the Google account you're signed into Sheets with has access to the Search Console property. (If not, ask the SC admin: Settings → Users and permissions → Add user at search.google.com/search-console.)
2. In GPTSheet → Menu → Connectors → Google Search Console → click Connect.
3. Approve the Google consent screen if prompted (one-time, shared with the other Google connectors).
4. The Site dropdown auto-populates with every Search Console property your Google account can access. Pick one — for domain properties this looks like sc-domain:example.com, for URL-prefix properties like https://www.example.com/. Save changes.
5. Done. Run actions: List queries, Top pages, Country breakdown, Custom report.

YouTube Insights

Channel + video performance via the YouTube Analytics API.

Setup

1. Make sure the Google account you're signed into Sheets with owns (or manages) the YouTube channel you want to query.
2. In GPTSheet → Menu → Connectors → YouTube Insights → click Connect.
3. Approve the Google consent screen if prompted.
4. Find your Channel ID: YouTube Studio → Settings → Channel → Advanced settings. It's a 24-character string starting with UC.
5. Paste the Channel ID → Save changes. Done.

YouTube Analytics only supports channels owned or managed by the signed-in Google account — there's no "grant access" model like GA4. If you don't see data, double-check that the signed-in account has channel access in YouTube Studio.

X (Twitter) Insights

Profile metrics, recent tweets, mentions, search.

Heads up on X API pricing. X (Twitter) made their API paid in 2023. Tiers as of 2026:

Set expectations: most GPTSheet users on the free tier will hit the cap in 1-2 scheduled refreshes. Plan for Basic if you want real recurring data.

Two token types — pick one

X gives you two ways to generate a token, and the connector accepts either. Pick based on what you need:

Setup — App-only Bearer Token (easiest)

1. Go to X Developer Portal → sign up if needed.
2. Create a Project → create an App in it (name them anything, e.g. gptsheet-personal).
3. App dashboard → Keys and tokens tab → Bearer Token section → click Generate. Copy immediately — X shows it once.
4. In GPTSheet → Menu → Connectors → X Insights:
   • Access token: paste the Bearer Token.
   • Username: your handle without the @ (e.g. fintwit_ai).
   • The X account dropdown auto-populates. Pick it.
5. Click Connect → verification returns your profile.

Setup — User Access Token (more metrics)

1. Same first two steps as above (create app in dev portal).
2. In the app dashboard, set User authentication settings:
   • App permissions: at minimum Read
   • Type of App: Web App, Automated App or Bot
   • Callback URI: https://example.com (placeholder is fine — we don't use the OAuth redirect)
   • Website URL: anything (your homepage works)
   Save.
3. Back in Keys and tokens → Access Token and Secret → click Generate. You'll see an OAuth 1.0a Access Token + Secret. For GPTSheet, we want the OAuth 2.0 token — look for an OAuth 2.0 Client section or use the "Generate token" tool that exposes the OAuth 2.0 User token.
4. In GPTSheet:
   • Access token: paste the OAuth 2.0 User token.
   • Username: leave blank — we'll call /users/me to identify you.
   • The X account dropdown auto-populates from /users/me.
5. Click Connect.

Behind the scenes: the X account dropdown first calls /2/users/me. If that returns 401 (you pasted an App Bearer Token), it falls back to /2/users/by/username/{username}. Either way, the numeric user ID gets stored and used by every action in this connector. You don't pick auth mode anywhere — paste the token, fill in username only if needed, and the connector does the right thing.

Meta Graph API setup (for Facebook + Instagram Insights)

Both Facebook Insights and Instagram Insights use Meta's Graph API. The auth model is the same: a Meta developer app + a long-lived access token. Set this up once and you can use it for both connectors.

Quick primer — Meta's 4 token types

Meta's access tokens come in different flavors. You'll only generate one of these in this guide, but it helps to know what's what:

For Facebook + Instagram Insights, follow Steps 1–4 below to get a long-lived User token. Paste it into the connector and GPTSheet swaps it to a never-expiring Page token automatically.

Step 1 — Create a Meta developer app (one-time, 3 min)

1. Go to developers.facebook.com/apps and click Create App.
2. App details page → name it anything (gptsheet-personal works) → contact email is your own → click Next.
3. Use cases page (Meta's app creation wizard) → pick Other. (The wizard offers preset bundles like "Authenticate users" or "Marketing API." Picking Other gives you a blank app where you can enable exactly the use cases you want in Step 2.)
4. App type page → pick Business → Next.
5. Confirm + Create. Meta drops you on the app dashboard.
6. Note the App ID (shown at the top of every page) and find App Secret under App settings → Basic (click "Show"). You'll need both in Step 4.

Step 2 — Configure use cases (mandatory)

Meta replaced the old "Add product" model with Use cases. Your new app already has 4 use cases pre-populated, but their permissions are off by default. You need to enable each permission you'll use, otherwise Graph API Explorer will show "No configurations available for user access token."

1. In the app dashboard left sidebar, click Use cases. You should see 4 cards: "Create & manage ads with Marketing API", "Create & manage app ads with Meta Ads Manager", "Manage everything on your Page", "Manage messaging & content on Instagram".
2. For Facebook Insights → click Customize on Manage everything on your Page → Permissions tab → tick all of:
   • pages_show_list
   • pages_read_engagement
   • pages_read_user_content
   • read_insights — required for Post insights, Video insights, Page fan growth, and Page video views. Without it those actions silently return no rows.
   • business_management
3. For Instagram Insights → click Customize on Manage messaging & content on Instagram → Permissions tab → tick:
   • instagram_basic
   • instagram_manage_insights
   • pages_show_list (Meta requires this even for IG — IG accounts attach to Pages)
4. Skip Create & manage ads with Marketing API and Create & manage app ads with Meta Ads Manager — those are for ad campaigns (not currently in GPTSheet's MVP).

Don't submit for App Review. Each permission shows "Standard access" by default. That's sufficient because you (as the app developer/admin) can generate tokens for yourself with Standard access. Advanced access (which needs App Review, ~2-6 weeks) is only required if you want strangers to OAuth through your app. Since GPTSheet is BYOK — each user creates their own Meta app for their own data — Standard access is always enough. Ignore any "App Review required" prompts.

Step 3 — Get a short-lived access token (1 min)

1. Open Graph API Explorer. Hard refresh (Cmd+Shift+R / Ctrl+Shift+R) if you had it open before — Meta caches the "no configs" state aggressively.
2. Top-right Meta App dropdown → select your app.
3. Click Get Token → Get User Access Token. The permissions checkboxes will reflect everything you ticked in Step 2.
4. Tick the permissions you enabled (same list as Step 2).
5. Click Generate Access Token → log in → approve. The token appears in the text box.

This token expires in about 1 hour. Convert it to a long-lived one in Step 4 before using it in GPTSheet.

Step 4 — Convert to a long-lived token (60-day validity)

Paste this URL into your browser (replace the three values):

https://graph.facebook.com/v19.0/oauth/access_token?
  grant_type=fb_exchange_token
  &client_id=YOUR_APP_ID
  &client_secret=YOUR_APP_SECRET
  &fb_exchange_token=YOUR_SHORT_LIVED_TOKEN

(Strip the line breaks; that's one URL on one line.) You'll get JSON back with a new access_token — this one lasts 60 days.

Step 5 (skip for FB / IG — only needed for special cases) — Get a never-expiring Page Access Token manually

You probably don't need this step. When you paste your long-lived User token into the Facebook Insights or Instagram Insights connector, GPTSheet's auto-discovery dropdown calls /me/accounts for you, lists your Pages, and on selection automatically swaps the token to that Page's never-expiring Page Access Token. So the never-expiring token derivation happens silently inside the connector.

This step is only relevant if:

• You want a Page token before opening GPTSheet (e.g. to test API calls in Postman / curl)
• You're scripting against the Meta Graph API outside of GPTSheet and want a Page token for your own use

To derive a Page token manually, paste this in your browser:

https://graph.facebook.com/v19.0/me/accounts?access_token=YOUR_LONG_LIVED_USER_TOKEN

The response lists every Page you manage. Each row has an access_token field — that's the never-expiring Page Access Token for that Page.

Facebook Page Insights

Page-level fan growth, post performance, audience demographics.

Setup

1. Complete the Meta Graph API setup above to get your long-lived User access token (Step 4).
2. In GPTSheet → Menu → Connectors → Facebook Insights: paste the User token into Access token.
3. The Page dropdown auto-populates with every Page you manage. Pick one — GPTSheet automatically:
   • Fills in the Page ID
   • Swaps the Access token to that Page's never-expiring Page Access Token (so scheduled refreshes never break)
4. Click Connect. You should see ✓ Connected.

Already have a Page token? Paste it in Access token; the Page dropdown will fall back to a text input. Type the Page ID manually.

Multiple Pages? Each Page can be added as a separate account. Click + Add another account at the top of the connector, paste your User token again, pick a different Page from the dropdown.

Available actions

• Page details — name, category, fan count, talking about count
• Recent posts — last 100 posts with reach, engagement, reactions, shares
• Post insights — per-post detailed metrics by post ID
• Page insights (date range) — daily impressions, reach, engaged users, page views
• Fan demographics — by country, city, locale, gender+age (requires ≥100 fans)

Instagram Insights

Business-account level + per-post + per-reel metrics. Audience demographics. Hashtag tracking.

Prerequisites

1. Your Instagram must be a Business or Creator account, not Personal. Switch in IG app → Settings → Account → Switch to Professional Account → Business.
2. Your IG account must be linked to a Facebook Page. IG app → Settings → Account → Linked Accounts → Facebook → connect the Page. (Create a Page first at facebook.com/pages/create if you don't have one.)

Setup

1. Complete the Meta Graph API setup above to get your long-lived User access token (Step 4). The same app + token covers both FB and IG.
2. In GPTSheet → Menu → Connectors → Instagram Insights: paste the User token into Access token.
3. The IG Business account dropdown auto-populates with every IG Business account linked to a Page you manage. Pick one — GPTSheet automatically:
   • Fills in the IG account ID (17841...)
   • Swaps the Access token to that Page's never-expiring Page Access Token
4. Click Connect. Verification fetches your username + follower count.

Don't see your IG account in the dropdown? Either (a) IG isn't linked to a Page in your Meta Business Suite, (b) the User token is missing instagram_basic scope — re-do Step 3 of Meta setup with that scope ticked, or (c) you're using a Page token already, in which case the dropdown falls back to a text input — paste the IG account ID manually.

Multiple IG accounts? Use + Add another account at the top of the connector to add each as its own auth record.

Available actions

• Account details — username, bio, profile picture, follower/follows/media counts
• Recent media — last 100 posts/reels/stories with engagement counts via media fields
• Media insights (per post) — reach, saves, total interactions, views, shares (Meta retired impressions/engagement at the insights endpoint in 2024)
• Reel insights (per reel) — reach, saves, total interactions, views, shares, follows attributed, profile visits driven
• Account insights (date range) — daily reach, views, total interactions, profile views, website clicks (results auto-pivot to one row per day)
• Follower count (30d) — daily new followers for the last 30 days
• Follower demographics (by country) — country breakdown. Use breakdown=city, gender_age, or age in the URL for other splits. Requires ≥100 followers
• Stories — currently-live stories
• Hashtag search — resolve hashtag string to ID for hashtag tracking

Common gotchas

• "(#100) Object does not support this operation" — your IG account isn't Business/Creator, or isn't linked to a Page.
• "Invalid OAuth access token" — token expired or lacks the right permissions. If you used the auto-discovery dropdown, just re-open the connector → pick the Page again to refresh. If you pasted a Page token manually, re-run Step 5 of Meta setup.
• Empty audience demographics — Meta withholds demographic data unless you have ≥100 followers.
• Graph API version — GPTSheet uses v19.0, supported by Meta through mid-2026. We'll bump it before deprecation.

PostHog

1. PostHog → Settings → Account → Personal API Keys → Create. Grant project:read scope (and any others you want).
2. Paste the key into GPTSheet — the Project field auto-populates with a dropdown of projects you can access (no need to manually look up the Project ID).
3. Host is usually https://us.posthog.com (or https://eu.posthog.com for EU).
4. Paste all three → Connect.

Amplitude

1. Amplitude → Settings → Projects → your project → API Keys.
2. Copy API Key and Secret Key.
3. Paste both → Connect.

Ghost

1. Ghost Admin → Settings → Integrations → Add custom integration.
2. Copy the API URL shown on the integration page (e.g. https://yourblog.ghost.io) — that's the Site URL we ask for.
3. Copy the Admin API Key from the same page. It looks like 6a1e5be944a3670001ab3704:02a287ee138878... — a 24-char id, a colon, then a 64-char secret. Paste it whole.
4. Paste both → Connect.

Notion

1. Go to notion.so/my-integrations → New integration.
2. Name it, copy the Internal Integration Token (starts with secret_).
3. In each Notion database or page you want GPTSheet to read, click … → Add connections → pick your integration.
4. Paste the token → Connect.

Airtable

Pull records, search by formula, get base + table schema. Bases + tables + IDs are discoverable in-app via the "Get base schema" action.

Setup

1. Go to airtable.com/create/tokens → Create new token.
2. Name it gptsheet.
3. Under Scopes, tick both:
   • data.records:read — required for List records, Get a record, etc.
   • schema.bases:read — required for List bases, Get base schema. Don't skip this — schema-discovery actions will return HTTP 403 without it.
4. Under Access, either pick specific bases or select "All current and future bases" if you want one token to work across everything.
5. Click Create token → copy the value (starts with pat; shown only once).
6. In GPTSheet → Connectors → Airtable → paste the token → Connect.

Finding base + table IDs

After connecting, run the List bases action — it returns every base your token can see with its appXXXXXXXXXX ID. Then run Get base schema with that ID to see every table's name + ID + field schema. Use these IDs in subsequent record-fetch actions.

Already created a token without schema.bases:read?

You don't need a new token — Airtable lets you edit existing tokens. Open airtable.com/create/tokens → click your token → add the scope → Save. The existing token immediately gains the new permission; no need to re-paste anywhere.

MySQL

1. You need a publicly-reachable MySQL host (RDS, Cloud SQL, PlanetScale, your own VPS). Apps Script cannot reach hosts in private VPCs without a tunnel.
2. Create a read-only DB user: CREATE USER 'gptsheet'@'%' IDENTIFIED BY '...'; GRANT SELECT ON yourdb.* TO 'gptsheet'@'%';
3. Allow Google Apps Script's outbound IP ranges through your security group (docs).
4. Paste host + port (3306) + database + user + password → Connect.

MS SQL Server

1. Public-reachable Azure SQL or self-hosted SQL Server.
2. Create a read-only login + map to a DB user with SELECT permissions.
3. Allow Apps Script outbound IPs through firewall.
4. Paste host + port (1433) + database + user + password → Connect.

Postgres

Direct JDBC connection to any publicly-reachable Postgres — RDS, Cloud SQL, Neon, DigitalOcean Managed Postgres, or self-hosted on a VPS.

Setup

1. You need a publicly-reachable Postgres host. Apps Script cannot reach hosts in private VPCs without a tunnel.
2. Create a read-only DB user:

   CREATE USER gptsheet WITH PASSWORD '...';
   GRANT CONNECT ON DATABASE yourdb TO gptsheet;
   GRANT USAGE ON SCHEMA public TO gptsheet;
   GRANT SELECT ON ALL TABLES IN SCHEMA public TO gptsheet;
   ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT SELECT ON TABLES TO gptsheet;

3. Allow Apps Script's outbound IP ranges through your security group (Apps Script JDBC IPs). For RDS: add the documented ranges to the inbound rule for port 5432.
4. In GPTSheet → Connectors → Postgres: paste host + port (default 5432) + database + user + password → Connect.

Hosted Postgres specifics

• AWS RDS: ensure the instance has Public access = Yes. Add Apps Script's IP ranges (or temporarily 0.0.0.0/0 for a quick test) to the security group's inbound port 5432.
• Google Cloud SQL: enable a public IP on the instance; add Apps Script's IPs as authorized networks.
• Neon: the Neon Postgres endpoint is publicly reachable by default. Copy host + database + user + password from the dashboard. Use require SSL mode (Neon enforces TLS).
• Supabase Postgres direct: use the Connection Pooler endpoint (port 5432) from Project Settings → Database. The host looks like aws-0-us-east-1.pooler.supabase.com. Or use the dedicated Supabase connector if you'd rather go through PostgREST + service_role.

If verification times out after 20 seconds, that's almost always a firewall — the connection never reaches Postgres. Double-check your security group inbound rules for port 5432, and that the host is reachable from the public internet.

Supabase

1. Supabase Project Settings → API Keys.
2. Click the "Legacy anon, service_role API keys" tab (the newer Publishable / Secret keys are not yet supported).
3. Copy the service_role secret value — it's the second row, labeled secret in red.
4. Grab your Project URL from the address bar or Project Settings (format: https://<ref>.supabase.co).
5. Paste both into GPTSheet → Connect.

The service_role key bypasses Row Level Security. Treat it like a database password — never expose it in client code. The Supabase docs recommend rotating it immediately if leaked.

DynamoDB

1. AWS Console → IAM → Users → Create user. Name it gptsheet-dynamo (or anything you like). Skip the "Provide user access to console" box.
2. On the next screen ("Set permissions"), pick Attach policies directly → Create policy (opens a new tab). Switch to the JSON tab and paste:

   {
     "Version": "2012-10-17",
     "Statement": [{
       "Effect": "Allow",
       "Action": [
         "dynamodb:ListTables",
         "dynamodb:DescribeTable",
         "dynamodb:Scan",
         "dynamodb:Query",
         "dynamodb:GetItem",
         "dynamodb:BatchGetItem",
         "dynamodb:ExecuteStatement",
         "dynamodb:PartiQLSelect"
       ],
       "Resource": "*"
     }]
   }

   Name it gptsheet-dynamodb-readonly → Create policy. Back on the user-creation tab, refresh the policy list, tick it, → Next → Create user.
3. Open the new user → Security credentials tab → Create access key → choose "Third-party service" → Next → Create. Copy the Access key ID and the Secret access key immediately — the secret is shown only once.
4. Find the AWS region where your tables live (e.g. us-east-1, us-east-2, eu-west-1). DynamoDB is regional — keys created in your account work in every region, but the connector needs to know which region to query.
5. In GPTSheet → Connectors → DynamoDB: paste region + access key ID + secret → Connect. Leave Session token blank unless you're using STS temporary credentials.

The granular PartiQLSelect action is what the AI SQL Builder and Custom PARTIQL Query box use — it's distinct from ExecuteStatement. Without it you'll see "not authorized to perform: dynamodb:PartiQLSelect". The policy above grants read-only access; if you want GPTSheet to also write data, add PartiQLInsert, PartiQLUpdate, PartiQLDelete, PutItem, UpdateItem, DeleteItem — but we recommend keeping it read-only.

Pricing

One license = one account.

Privacy

GPTSheet is built around a local-first data flow:

Read our full privacy policy.

Updates

Basic users get all updates released within 1 year of purchase. After that, your existing version keeps working forever — you just don't get new formulas or provider integrations released after the cutoff. Upgrade to Premium any time for lifetime updates.

Premium users get every update, forever. New formulas, new providers, new sidebar features — all included.

Refunds

All sales are final. If you run into a bug or activation problem, email support@gptsheet.com and we'll work with you to fix it.

FAQ

Why is GPTSheet cheaper than GPT for Work?

GPT for Work bundles the LLM cost into expiring credit packs. We don't — you pay providers directly. For a moderate user, that means $1–3/month in LLM costs vs $29–$299 in expiring GPT for Work credits.

Do you train on my data?

No — our server never sees your sheet data, prompts, or LLM responses. Your data flows directly to the LLM provider you chose; how they handle it is governed by their privacy policy (OpenAI, Anthropic, and Google all support API opt-out from training).

Can I use a local model with Ollama?

Premium tier supports custom OpenAI-compatible endpoints. Point at http://localhost:11434/v1 and run any model you've pulled. Useful when you want zero outbound LLM calls.

Does it work in Excel?

Yes. Same formulas, same sidebar UI, same chat agent. One license covers both Google Sheets and Microsoft Excel.

What happens after my 1-year Basic updates window ends?

Your existing version keeps working forever. You stop receiving new features released after the 1-year mark. Upgrade to Premium any time for lifetime updates, or stay on your current version indefinitely.

Troubleshooting

"Google hasn't verified this app" warning

Expected for newly-published add-ons before Google completes OAuth verification (usually 2-6 weeks after submission). Once verified, the warning disappears for all users.

To proceed in the meantime: click Advanced → Go to GPTSheet (unsafe) → Allow. The "unsafe" label is generic — it does not mean the add-on is malicious; only that the verification review is still pending.

HTTP 429 from OpenAI when generating images

Cloudflare rate limit triggered by OpenAI's DALL·E 3 service on free / Tier 0 accounts. Switch to a Gemini image model:

=IMAGE(AI_IMAGE("a green apple", "", "gemini-2.5-flash-image"))

"Authorization required" / "Specified permissions are not sufficient"

Apps Script triggers re-authorization when scopes change (e.g. we added Drive access for image generation). To re-grant:

1. Open Extensions → Apps Script from your sheet.
2. Pick any function in the toolbar and click Run.
3. Approve the re-authorization dialog. New scopes are added to your existing grant.

Cell shows [gptsheet] OpenAI key not set

The cell formula needs its key synced to Apps Script. Open the sidebar → ≡ → API keys → edit OpenAI → Save. (Saving also syncs to formulas.)

Need more help?

Email support@gptsheet.app. Premium customers get priority response (typically within 24h).