GRID
A headless spreadsheet engine that software teams embed in their own products and AI agents. It runs Excel and Google Sheets workbooks natively and lets agents read, edit and recalculate them through structured tool calls instead of generated code.
What is GRID?
GRID is a spreadsheet engine that software teams embed inside their own products and AI agents. It ships as npm packages, @grid-is/spreadsheet-engine alongside spreadsheet-viewer, spreadsheet-editor and agent-tools, rather than as an application you log into.
The problem it targets is narrow and real. When a language model has to read or change a spreadsheet it normally writes throwaway code to do it. That code is slow, costly in tokens and grows less reliable as the workbook gets more complex. GRID replaces code generation with a structured interface: the agent calls an operation to read a cell, apply a change or recalculate, and the engine answers deterministically.
The engine is a headless TypeScript library built around a dependency graph. When a value changes it walks the graph and recomputes only the cells that depend on it, which GRID calls animation-speed recalculation. Compatibility is the other pillar: existing Excel and Google Sheets workbooks run unmodified, both spilling models and both cell-metadata conventions are supported, and close to 200,000 automated unit tests run on every build to keep it that way. The library runs wherever the host product runs, from the browser to the user's device, a Node.js server or the customer's own cloud, so workbook data need not leave that infrastructure.
Agent Tools packages the same engine as 27 described, schema-validated tools grouped into Understand, Audit, Edit and Model. They cover workbook structure and search, precedent and dependent tracing, formula-error listing, cell and style edits that return a delta of what changed, and modelling operations including what-if scenarios and goal seek. The package runs as an MCP server or drops into an agent SDK. On SpreadsheetBench's expert-verified 400, the benchmark team scored GRID's tools at 91.25% running Claude Sonnet.
Two ready-made interface surfaces sit on the same engine, a themeable viewer for displaying workbooks and an editor for letting users change them. A white-label Excel add-in is also offered, where the engine sandboxes the file so a model's mistakes never reach it.
GRID ehf. is based in Reykjavík, Iceland, was founded in 2018 and has raised roughly USD 16.5 million.
What it does
- Recalculate Excel and Google Sheets workbooks natively inside your own application
- Give an AI agent 27 tools to read, audit, edit and model a workbook through MCP or your agent SDK
- Run goal seek and what-if scenarios in a single function call instead of generated code
- Generate whole spreadsheets from scratch with a language model
- Display workbooks in your product with a themeable, client-side viewer
- Let users edit spreadsheets without leaving your interface
- Build a white-label Excel add-in with the engine sandboxing the file
When to use GRID / When not to
A quick filter to help you decide if GRID is the right fit.
When to use GRID
- Product and engineering teams embedding spreadsheet calculation inside a SaaS or internal application
- Developers building AI agents that must read, audit and edit real workbooks reliably
- Companies whose numbers are contractual, such as the investment-banking platform GRID describes as a customer
- Teams sitting on established Excel models they need to keep alive without rewriting the logic in code
- Products with strict data-locality requirements, since the engine runs in the browser, on-device or in your own cloud
When not to use GRID
- Individuals looking for a spreadsheet to open and use: there is no app, no account and nothing to sign up for
- Anyone hoping to run a commercial product or a revenue-generating internal workflow on the free licence, which forbids exactly that
- Teams that need production-grade guarantees today, since the evaluation licence disclaims warranties, support and any use on critical data
- Buyers who need to budget before talking to a vendor, because no commercial price is published anywhere
- Mobile-first use cases: there is no iOS app, no Android app and no browser extension
How to use GRID
A typical end-to-end flow, from setup to results.
- Decide which surface you need: the headless engine for calculation, the viewer for display, the editor for editing, or agent-tools for an AI agent
- Install from npm, for example npm install @grid-is/spreadsheet-engine, with no account, sign-up or API key required
- Alternatively hand the job to a coding agent: paste GRID's prompt into Claude Code, Cursor or Copilot and it runs npx -y skills add GRID-is/skills
- To wire an agent at runtime, start the MCP server with npx -y @grid-is/agent-tools, or import createGridTools() into your own SDK
- Load a workbook in code with Model.fromJSF(spreadsheet), then write and read values with model.write() and model.readValue()
- Add further workbooks with model.addWorkbook() and cross-reference them using the [budget.xlsx]Sheet1!A1 syntax
- For a React front end, render the SpreadsheetViewer component from @grid-is/mondrian-react, passing it your model and your own theme object
- Follow the per-package quickstarts on docs.grid.is and the API reference at docs.grid.is/apiary
- Keep the visible 'Powered by GRID' attribution in place for as long as you are on the evaluation licence
- Before you ship commercially, contact GRID through the licensing form to scope and price a commercial licence
Pros & Cons
Pros
- Results are deterministic: the figure returned is the one the spreadsheet produces, not the model's guess
- 91.25% on SpreadsheetBench's expert-verified 400, scored by the benchmark team rather than self-reported
- Excel and Google Sheets compatibility is backed by close to 200,000 unit tests on every build
- Spreadsheet data stays in your own infrastructure, with no cloud round-trips
- Installs from npm with no account, sign-up or approval, and the evaluation licence is genuinely full-featured
- One engine behind three coherent surfaces, so the headless library, viewer and editor stay consistent
- The engine has been in production since 2021 behind Calculator Studio, so it predates the current agent wave
Cons
- No public pricing at all: the commercial licence is quoted case by case, so you cannot compare or budget alone
- The free licence bars every commercial use, including an internal workflow that merely saves money
- Evaluation users must display a visible 'Powered by GRID' attribution and get no support obligation, with liability capped at USD 100
- Reference customers are anonymised as 'major investment banks' and 'a leading LLM provider', so nothing is verifiable
- The privacy policy is visibly dated, describing an alpha-stage GRID API Service with no effective date or version
- No DPA, no subprocessor list, no SOC 2 or ISO 27001, and no trust or security page
- Developer-only, English-only, with no mobile app and no browser extension
Pricing & Plans
GRID operates a two-tier licensing model with no published price point. A free evaluation licence is bundled with every npm package, requires no sign-up and carries no time limit, but is restricted to development, testing, internal prototypes and personal, educational or other non-commercial use. Any commercial use, whether a shipped product or an internal workflow that generates revenue or cost savings, requires a commercial licence, which GRID states is priced according to your use case. No amount, currency or billing unit is disclosed on the licensing page, the get-started page, the product FAQs or anywhere else on the site, so the entry cost cannot be established without contacting the company.
- Evaluation License — free and fully featured
- bundled with every npm package and requiring no sign-up. Limited to development
- testing
- internal prototypes and personal
- educational or non-commercial use. Requires a visible 'Powered by GRID' attribution
- includes anonymous telemetry
- and carries no support obligation.
- Commercial License — price on request
- scoped to your use case. Required as soon as GRID runs for a commercial entity
- in a product you ship or an internal workflow you operate. Comes with direct support from the GRID team and drops both the attribution requirement and the telemetry.
Data, GDPR & hosting
A consolidated view of how GRID handles your data.
GDPR overview
The privacy policy states a commitment to the data subject rights set out in the EU's GDPR and lists them: access, correction, amendment, deletion, portability, restriction of processing, objection, and complaint to a data protection authority. Transfers to providers outside the EU/EEA are said to rely on the European Commission's standard data protection clauses. Beyond that the implementation is thin. No data protection officer is named, no Article 27 EU representative is designated, no data processing agreement is published or offered, and no subprocessor list exists even though third-party providers are acknowledged. The policy carries neither an effective date nor a version number and still describes a GRID API Service in alpha, which no longer matches the npm packages the company ships. GRID ehf. is established in Iceland, inside the EEA but outside the European Union.
Who owns the data?
GRID separates two situations. For account data tied to its GRID API Service the company acts as controller and says it discloses personal data to no one unless required by law or a court order. For files you upload the roles reverse: you or your company are the data controller and GRID is only the processor, holding no controller role over third parties named inside those files. Ownership of the software never moves, as GRID keeps all intellectual property and grants no rights beyond the licence itself. One asymmetry is worth noting: any feedback you send GRID becomes theirs to use, modify and ship perpetually, worldwide and free of charge, with no obligation to you.
Reuse rights
Under the evaluation licence you may freely publish whatever the software produces, such as rendered spreadsheets and calculation results, and you may build, publish and share non-commercial projects, open-source ones included, along with blog posts, tutorials and case studies about using it. What you may not reuse is the software's internals: decompiled output, internal data structures, performance metrics and any analysis meant to expose how it works are all off limits, and that restriction survives termination. Reusing the software commercially requires a paid licence. The data you feed the engine stays yours, and with the npm packages it never leaves your infrastructure: GRID's only outbound signal is an anonymous initialisation ping carrying the package name, version and runtime environment, with no spreadsheet content and no personal data.
Data retention & training
Hosting summary
GRID names no hosting country and no hosting region anywhere on its site, and for the main product that silence is structural rather than an omission. The engine, viewer, editor and agent tools are libraries that execute inside the customer's own environment, whether that is a browser, an end-user device, a Node.js server or a private cloud, so workbook content is hosted by whoever deploys them. GRID states plainly that there are no cloud round-trips and that spreadsheet data never leaves your infrastructure. What GRID itself holds is much narrower: account data for its GRID API Service, meaning email address, IP address and system or diagnostic data, plus the anonymous initialisation ping sent by the npm packages. The privacy policy acknowledges third-party providers, some outside the EU/EEA, with transfers covered by standard contractual clauses, but names none of them. The grid.is site resolves to a Cloudflare anycast address, which reflects a CDN edge and says nothing about where data is stored. Jurisdiction is Icelandic: the licence is governed by Icelandic law with exclusive venue in the District Court of Reykjavík.
Things to keep in mind
Risks and trade-offs to weigh before adopting GRID.
- The definition of commercial use is deliberately broad, covering any workflow that generates revenue or cost savings, including internal ones and work done by AI agents or contractors on your behalf
- GRID may terminate the evaluation licence at any time, with or without cause, and total liability is capped at USD 100
- There are no warranties and no support obligation on the evaluation licence, so a production dependency built on it is a business risk
- Any feedback you send is licensed to GRID perpetually, worldwide and free of charge, which is easy to overlook when reporting bugs
- Telemetry is on by default in evaluation packages; it is narrow, but you should still declare it in your own privacy documentation
- The privacy policy describes an alpha-stage service and shows two different company addresses across the site, so treat its legal detail as needing confirmation
- Icelandic law and the exclusive jurisdiction of the District Court of Reykjavík apply, which can make disputes costly for a distant licensee
Setup & Integrations
Technical difficulty
Straightforward for a developer, impossible without one. Installation is a single npm command with no account, no sign-up and no API key, since the evaluation licence ships inside every package. A few lines load a workbook and read calculated values, and GRID claims most developers are running within minutes; a coding agent can do it in one command using the official skills. The real work is not installation but integration, deciding how spreadsheet logic fits your product, and then negotiating a commercial licence before you ship. One cited LLM provider deployed the viewer within days.
Deployment
Integrations
Behind GRID
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement GRID.
Frequently asked questions
What exactly is GRID?
Do I need an account to try it?
When do I have to pay, and how much?
Does the free licence cover internal business use?
Will it work with my existing Excel files?
Do my spreadsheets get sent to GRID?
How do I connect GRID to an AI agent?
Does the engine modify my original file?
Does GRID publish a DPA or security certifications?
Who is behind GRID?
Should you pick GRID?
GRID is infrastructure, not a finished product, and it should be judged as such. It does one thing and states it plainly: make spreadsheet calculation reliable for software and for AI agents, by letting them call operations instead of writing code. The technical case is unusually well evidenced for this category. A 91.25% score on SpreadsheetBench was produced by the benchmark team rather than by GRID, the compatibility claim rests on almost 200,000 unit tests per build, and the engine has been running in production since 2021 behind the company's own Calculator Studio, so it predates the agent wave it now serves.
The commercial side is where you should slow down. Nothing is priced publicly, the free licence excludes any commercial use including internal workflows that merely save money, and the two flagship customers are described only as major investment banks and a leading LLM provider. The legal and documentary layer also lags the product: the privacy policy still describes an alpha-stage API service, carries no effective date, and sits alongside no DPA, no subprocessor list and no certifications. For a component that will sit inside someone else's product, that gap matters more than it would for a consumer tool.
For a team that already knows it needs spreadsheet logic in software, GRID is worth an afternoon: the install costs nothing, needs no account, and the evaluation licence is fully featured. Just plan for a sales conversation before anything ships, and ask directly for the security and data processing documents the site does not publish.
- Choosing a selection results in a full page refresh.
- Opens in a new window.