GPT for Work
An AI agent that lives inside Microsoft Excel and Google Sheets. It writes formulas, formats cells, builds pivot tables and charts, then cleans, translates, categorises and enriches data row by row across hundreds of thousands of rows.
What is GPT for Work?
GPT for Work is an AI agent that runs inside Microsoft Excel and Google Sheets, built by the Luxembourg software company Talarian. It adds a chat panel to the spreadsheet: you describe what you want in plain language, the agent reads the file, plans the steps and executes them in place. There is no configuration interface to learn and, in the vendor's words, no prompt engineering required.
Three layers sit under that chat panel. The Agent handles conversational requests such as writing a formula, correcting an error, formatting a range, building a pivot table or drawing a chart. The bulk tools apply a prompt template to every row of a table, using subagents to run those prompts in parallel: the agent drafts one template per column and picks a suitable model itself. The GPT functions, written as =GPT() and its specialised variants, prompt the AI directly from a cell and can be dragged across thousands of cells, referencing other cells so that results react to changes.
The throughput claims are specific: up to 1,000 answers per minute depending on the model and task, up to 200,000 rows per run in Google Sheets and up to a million in Excel, with real-time progress tracking and a one-click stop. Typical bulk jobs shown on the site are translating product descriptions, scoring leads against written criteria, categorising catalogues, enriching company lists by web search and deduplicating contacts.
There is deliberately no model lock-in. OpenAI, Anthropic, Google, Perplexity and Parallel models are built in; DeepSeek, Mistral and xAI can be connected with your own key; local models can be reached through Ollama or any OpenAI-compatible endpoint. A companion service, Autosheet, exposes the Google Sheets agent over MCP and an HTTP API so that Claude, ChatGPT, Cursor or a script can drive a spreadsheet.
The product started in early 2023 as GPT for Sheets and Docs, gained an Excel and Word edition in mid-2023, and now claims more than 55,000 paying customers and 7 million installations.
What it does
- Write, check, fix and explain spreadsheet formulas
- Apply AI to every row in bulk: translate, categorise, generate, score, enrich
- Clean, deduplicate, standardise and validate messy data
- Build pivot tables, charts, dashboards and written analyses
- Format cells, including conditional formatting, across large sheets
- Run one web search per cell and extract information from unstructured text
- Analyse images in bulk from URLs using vision models
When to use GPT for Work / When not to
A quick filter to help you decide if GPT for Work is the right fit.
When to use GPT for Work
- Finance and accounting teams who close the books in spreadsheets and need to reconcile exports, validate formulas and build forecast models
- Sales and RevOps staff who clean pipeline data, enrich contact records and score leads in bulk
- E-commerce and SEO practitioners generating, translating and tagging catalogue content at catalogue scale
- Consultants and business analysts who restructure client datasets and turn them into decision-ready summaries
- HR, supply-chain and operations people maintaining large reference tables of employees, suppliers or stock
When not to use GPT for Work
- Anyone who needs the tool on a phone or tablet: there is no iOS or Android app, only desktop and web add-ins
- Teams under a data-residency obligation that keeps personal data inside the EU, since customer data is hosted only in the United States
- Excel users who want programmatic access, because the Autosheet API and MCP server cover Google Sheets only
- People looking for a permanently free tool: the free credits are a trial, after which every space moves to paid usage
- Organisations running old Office builds below the documented minimum versions for Windows or macOS
How to use GPT for Work
A typical end-to-end flow, from setup to results.
- Open the install page and pick your platform: the Microsoft Office add-in for Excel and Word, or the Google Workspace add-on for Sheets and Docs
- Install from the marketplace and grant the authorisation scopes the add-in requests; new users get a free trial with starting credits and no card required
- Sign in with Microsoft Entra ID or Google single sign-on
- Open the GPT for Work chat panel from the spreadsheet ribbon or sidebar
- Type what you need in plain language, for example asking for a pivot table by region and month, a translated column, or a clean-up of the current sheet
- For row-by-row work, let the agent draft a prompt template per column, then launch the run and watch progress in the sidebar, stopping or resuming in one click
- For reactive prompts, type a GPT function directly in a cell and drag it down across the range
- Optionally add your own provider API key or a custom OpenAI-compatible endpoint to route usage through your own account
- Administrators sign in to the web dashboard to create the team space, choose which models are available, set monthly spending limits and manage billing
- To drive Google Sheets from another AI client, connect the Autosheet MCP server to your agent and send it a prompt plus a spreadsheet URL
Pros & Cons
Pros
- Fast in the vendor's own benchmark of 12 March 2026: best result in ten of eleven tests, with an agent average of 16 seconds against 1 minute 57 for Copilot in Excel
- Genuine bulk capability, which none of the three competitors tested offers at comparable volume
- No model lock-in: several built-in providers, plus your own API key or a custom endpoint
- Works on both sides of the fence, in Excel and in Google Sheets, where the compared rivals are single-platform
- Strong security posture: ISO 27001, Google CASA Tier 3, Microsoft publisher attestation, encryption at rest and in transit
- Customer data is never used to train language models, and the vendor states it never sells data
- Spending is controllable: pooled team credits, monthly caps per team or user, and a one-click pause
Cons
- Customer data is hosted only in the United States, in the Google Cloud us-central1 region in Iowa, despite the vendor being based in the EU
- Without your own API key, Inputs and Outputs are logged for thirty days, kept up to a year for product improvement and may be stored for as long as five years
- The comparative benchmark that anchors the speed claims is published by the vendor, not by an independent third party
- The API and MCP server cover Google Sheets only, so Excel users get no programmatic access
- No mobile application and no permanently free plan
- Pay-as-you-go cost is hard to forecast: it varies with the model, the amount of text per row and the complexity of the task
- The Google bulk-tools scope grants access to all of the user's spreadsheets and cannot be narrowed, a Google API limitation the vendor documents openly
Pricing & Plans
There is no permanent free plan. New users receive a free trial with starting credits and no card required, after which the space moves to paid usage. Two paid routes exist. Pay-as-you-go requires a minimum credit purchase of 29 USD, with packs at 29, 79, 139 and 299 USD and upwards. Subscription pricing, currently rolling out to new spaces, starts at 25 USD per month for the individual Standard plan, which is the lowest recurring price point. Users supplying their own provider API key still pay a platform fee of 1 USD per million tokens.
- Pay-as-you-go Starter — 29 USD credit pack (minimum purchase)
- credits valid 12 months
- Pay-as-you-go Essential — 79 USD credit pack
- Pay-as-you-go Pro — 139 USD credit pack
- presented as the most popular
- Pay-as-you-go Scale — 299 USD and above
- Subscription Standard — 25 USD per month
- individual plan for one Google or Microsoft user
- 25 USD of monthly usage credits included
- Agent
- bulk tools
- GPT functions and Autosheet API access for Google Sheets
- Subscription Business — 45 USD per seat per month
- team plan
- 25 USD of monthly usage credits per seat
- everything in Standard plus team features
- Subscription Enterprise — custom pricing
- organisation plan
Data, GDPR & hosting
A consolidated view of how GPT for Work handles your data.
GDPR overview
GDPR compliance is claimed explicitly. Talarian says it acts as controller for its own processing under its privacy policy and as processor under a publicly available Data Processing Agreement, both last updated on 4 August 2026. A Data Protection Officer is designated by name in the privacy policy, with a postal address in Luxembourg, an email address and a phone number; a generic privacy contact is also published. The privacy policy lists the usual data-subject rights, including access, rectification, erasure, restriction, portability, objection and post-mortem directives, and states that transfers outside the EEA rely on an adequacy decision or on the European Commission's Standard Contractual Clauses. The company is established in Luxembourg, so no Article 27 representative is required or designated. Talarian is ISO 27001 certified, holds Google CASA Tier 3 certification and a Microsoft publisher attestation, and commissions annual third-party security assessments.
Who owns the data?
The customer keeps ownership. The terms of service state that, subject to the AI providers' own terms, the customer retains all right, title and interest in the data, files, attachments, text, images and personal information it puts into the service, and the security FAQ repeats that as between the user and Talarian the user retains all rights in both Inputs and Outputs. Talarian keeps ownership only of the service itself, its software and its trade marks, which the customer may not use. One exception cuts the other way: any feedback or suggestion sent to Talarian is irrevocably assigned to Talarian.
Reuse rights
Yes. Because the customer keeps every right in its Inputs and Outputs, the results produced in the spreadsheet can be reused, published, sold or fed into other systems without asking Talarian for permission. What Talarian itself does with that material depends on the configuration. With a custom endpoint, such as a local Ollama server or an Azure deployment, the data never touches Talarian's infrastructure at all. With the user's own API key, nothing is logged. Without an API key, Inputs and Outputs are logged for thirty days for support, kept up to a year for product improvement and may sit in storage for as long as five years. In no configuration is the material used to train language models, and Talarian states plainly that it never sells data.
Data retention & training
Hosting summary
Customer data is processed on Google Cloud Platform, in the us-central1 region in Iowa, United States. Encryption at rest uses Google's AES-256 mechanism, built on the Tink library and its FIPS 140-2 validated BoringCrypto module; data in transit travels only over TLS 1.2 or TLS 1.3. API keys, bulk tool configuration history and the GPT function cache are additionally encrypted in use. Physical data-centre security is Google's responsibility. The marketing website itself is hosted separately by Vercel, Inc. in California. Subprocessors are published: Google for hosting and generative AI, Zendesk for support, OpenAI, Anthropic and Perplexity for generative AI when used through the vendor's own keys, and Sentry for error monitoring. Separate processors handle CRM, transactional email, internal collaboration and analytics.
Things to keep in mind
Risks and trade-offs to weigh before adopting GPT for Work.
- Customer data is hosted only in the United States: check this against your own data-residency and transfer obligations before processing EU personal data
- If you do not supply your own API key, your prompts and results are logged and may be retained for up to five years
- The bulk tools require a Google scope covering all of your spreadsheets, which cannot be narrowed; you are relying on the vendor's commitment to touch only the files where the add-on was launched
- Pay-as-you-go spending scales with volume and model choice, so a single large run can burn a credit pack: set the monthly cap before the first bulk job
- The published speed comparison is the vendor's own benchmark, not an independent test
- Subscription pricing is still rolling out to selected spaces, so the advertised monthly rate may not be available to you
- Letting an agent rewrite, reformat and deduplicate a spreadsheet in place makes errors easy to miss: keep a copy of the source file and spot-check results rather than trusting a run wholesale
Setup & Integrations
Technical difficulty
Low. Installation is a one-click add-in from the Google Workspace Marketplace or the Microsoft marketplace, followed by single sign-on with Google or Microsoft Entra ID. No prompt engineering is required: the agent drafts its own prompt templates and picks the model. The only real prerequisites are a recent enough Office build on Windows or macOS, and, for teams, a few minutes in the web dashboard to invite users, choose permitted models and set spending limits. Connecting your own API key or a custom endpoint is optional and the one genuinely technical step.
Deployment
Integrations
Supported languages
Behind GPT for Work
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement GPT for Work.
Frequently asked questions
Where does GPT for Work actually run?
Is there a free plan?
How much does it cost?
Can I use my own AI provider API key?
Which AI models are available?
Is my data used to train AI models?
Where is my data hosted?
How long is my data kept?
Is there an API?
How many rows can it process?
Should you pick GPT for Work?
GPT for Work is a mature product rather than an experiment. Launched in early 2023, it claims more than 55,000 paying customers and 7 million installations, and it comes from Talarian, a privately held company that says it has been profitable since 2015, employs around 40 people and holds ISO 27001 certification. That matters for a tool you would trust with a finance close or a product catalogue.
Its real distinction is bulk work. Plenty of assistants can write one formula; far fewer will apply an instruction to two hundred thousand rows, retry the failures, put the answers back in the right cells and let you watch and stop the run. Add genuine provider choice, including your own API key or a local endpoint, and support for Excel and Google Sheets alike, and the positioning is coherent.
The reservations are equally concrete. Customer data sits exclusively on Google Cloud in Iowa, which is a conversation to have with your DPO even though the vendor is Luxembourg-based and offers a public DPA. Without your own API key, prompts and results are logged and may be stored for years. The API and MCP server serve Google Sheets only. And the speed figures that anchor the marketing come from a benchmark the vendor ran itself, which is worth reading as such.
For a team that already lives in spreadsheets and has repetitive, high-volume work to get through, the free trial costs nothing to evaluate and the spending caps make a controlled pilot straightforward. For anyone needing EU data residency, mobile access or programmatic control of Excel, the fit is weaker.
- Choosing a selection results in a full page refresh.
- Opens in a new window.