bilig

<p align="center"> <img src="docs/assets/github-social-preview.png" alt="bilig headless spreadsheet engine for Node.js programs" /> </p>

bilig is a headless workbook runtime for TypeScript. Use it when the spreadsheet is the model, not the UI: pricing rules, finance checks, agent tools, import jobs, queue workers, and serverless routes that need formulas to recalculate the same way every run.

Project site: https://proompteng.github.io/bilig/

When It Fits

Good fit	Use something else when
A Node service needs workbook formulas, named sheets, edits, and readback	You only need a visual grid component
An agent tool must write an input cell and return the calculated result	You need desktop Excel automation or Office macros
A backend job needs to save and restore formula-backed workbook state	Your workflow starts and ends with manual XLSX editing
Tests need a deterministic workbook model instead of screenshots	You want a general-purpose table library with no spreadsheet formula model

Proof You Can Reproduce

• Run the 90-second TypeScript check below. It edits one input, restores the saved JSON document, and verifies the dependent formula result.
• Run pnpm workpaper:bench:competitive:check. The checked-in artifact shows 46/46 comparable WorkPaper mean wins and names the slower p95 row: lookup-approximate-duplicates at 1.043x.
• Read the compatibility limits before importing real Excel workbooks.
• Track public signals in the growth snapshot: stars, npm downloads, starter issues, Discussions, traffic, and clones.
• The WorkPaper MCP server is listed in the official MCP Registry and on Glama. The directory status page keeps the npm command and directory evidence in one place.

If the 90-second check matches a problem you have, star the repo so you can find it again: https://github.com/proompteng/bilig/stargazers.

Start Here

Job	Start with
Prove the npm package	@bilig/headless in 90 seconds and the npm-only smoke test page
Build a Node workflow	server-side spreadsheet automation, five runnable workbook automation examples, framework adapters, evaluate Excel formulas in Node.js, and the Node service recipe
Wire an agent or MCP tool	agent tool-calling recipe, MCP spreadsheet tool server guide, MCP directory status, MCP client setup, and Claude Desktop MCPB bundle
Compare alternatives	JavaScript spreadsheet library guide, headless engine comparison, HyperFormula comparison
Contribute	starter issues, GitHub Discussions, and CONTRIBUTING.md

Useful direct paths:

• Node workflow examples: invoice totals, budget variance, fulfillment capacity, quote approval, subscription MRR, and serverless API route. Use Express, Fastify, and Hono adapters when the route is moving into an existing service.
• Agent and MCP adapters: agent framework guide, framework-specific notes for Mastra, LlamaIndex.TS, LangGraph.js, CopilotKit, and Cloudflare Agents, framework adapters, MCP tool server shape, and MCP server guide. Check the MCP directory status page when a directory listing needs the canonical npm command or registry proof. Use MCP client setup for Claude, Cursor, VS Code, Cline, and Codex configs, or build the Claude Desktop MCPB bundle. Run npm run agent:framework-adapters, npm run agent:mcp-tools, npm run agent:mcp-stdio, or pnpm mcpb:workpaper:build, or npm exec --package @bilig/headless -- bilig-workpaper-mcp.
• Open feedback threads: Postgres, Redis, and object storage adapters discussion and JavaScript spreadsheet library guide discussion.
• SEO comparison pages: evaluate Excel formulas in Node.js, Node spreadsheet formula engine, JavaScript spreadsheet library for Node services, SheetJS and ExcelJS alternative, headless engine comparison, and HyperFormula comparison.

The published MCP Registry entry is io.github.proompteng/bilig-workpaper.

Try @bilig/headless in 90 seconds

The fastest evaluation path uses the published npm package only. It builds a formula-backed workbook, applies an edit, persists the document, restores it, and throws if formula readback does not survive the round trip.

mkdir bilig-headless-eval
cd bilig-headless-eval
npm init -y
npm pkg set type=module
npm install @bilig/headless
npm install -D tsx typescript @types/node
curl -fsSLo eval.ts https://proompteng.github.io/bilig/npm-eval.ts
npx tsx eval.ts

Expected output:

{
  "before": 24000,
  "after": 38400,
  "afterRestore": 38400,
  "sheets": ["Inputs", "Summary"],
  "bytes": 1000,
  "verified": true
}

The TypeScript file is maintained in examples/headless-workpaper/npm-eval.ts. The exact byte count can change between package versions; verified: true and matching after/afterRestore values are the check.

The maintained repository example adds agent-style writeback verification:

git clone https://github.com/proompteng/bilig.git
cd bilig/examples/headless-workpaper
npm install
npm start
npm run agent:tool-call
npm run agent:framework-adapters
npm run agent:verify

Expected proof from npm run agent:tool-call includes:

{
  "toolCall": { "toolName": "setInputCell" },
  "toolResult": {
    "editedCell": "Inputs!B3",
    "before": { "expectedArr": 60000, "targetGap": -34000 },
    "after": { "expectedArr": 96000, "targetGap": 5600 },
    "verified": {
      "formulasPersisted": true,
      "restoredMatchesAfter": true,
      "expectedArrImproved": true,
      "targetGapClosed": true
    }
  }
}

Expected proof from npm run agent:verify includes:

{
  "after": {
    "customers": 65,
    "grossMrr": 15600,
    "expansionMrr": 18720,
    "annualizedArr": 224640,
    "arrTargetDelta": 74640
  },
  "verified": {
    "formulasUnchanged": true,
    "formulasPersisted": true,
    "restoredMatchesAfter": true
  }
}

The serverless route example gives HTTP and agent-tool evaluators a runnable JSON boundary:

git clone https://github.com/proompteng/bilig.git
cd bilig/examples/serverless-workpaper-api
npm install
npm run next-route-handler
npm run smoke
npm run persistence-adapters

Expected proof from npm run next-route-handler, npm run smoke, and npm run persistence-adapters includes:

{
  "edit": {
    "records": 4,
    "after": {
      "totalRevenue": 48600,
      "westCustomers": 20,
      "largestDeal": 24000
    },
    "checks": {
      "totalRevenueChanged": true,
      "formulasPersisted": true
    }
  },
  "verified": true
}

It is not a table widget. The repo contains a real workbook engine, formula parser/compiler, React workbook reconciler, reusable grid shell, binary sync protocol, agent API, browser/server persistence layers, and a conservative AssemblyScript/WASM fast path for formula families that have proven parity.

The long-term target is a spreadsheet platform that can be edited by people or agents, restored locally, synchronized through ordered mutation streams, and benchmarked against serious spreadsheet-engine workloads.

Why Watch This Repo

• Spreadsheet engine, not just UI: workbook mutation, formulas, snapshots, history, selections, dependency inspection, replica hooks, and import/export live below the React shell.
• Local-first by design: browser sessions restore from local state, preserve replica snapshots, and keep outbound edits as replayable mutation batches.
• Agent-addressable workbooks: the engine exposes stable request, response, event, and subscription shapes so agents can operate on spreadsheet state without screen scraping.
• Performance tied to proof: formula acceleration and WorkPaper benchmark work are backed by parity fixtures, differential checks, counters, and CI gates instead of benchmark-only rewrites.
• Reusable package boundaries: formula, core, grid, renderer, transport, protocol, storage, benchmark, and runtime concerns are split into packages.

What Works Today

• Create, mutate, snapshot, restore, undo, redo, and inspect workbooks through @bilig/core.
• Parse, bind, compile, and evaluate spreadsheet formulas through @bilig/formula, with fixture-driven parity checks.
• Render and navigate a virtualized browser spreadsheet shell through apps/web and @bilig/grid.
• Author deterministic workbooks with React components through @bilig/renderer.
• Exercise the product runtime through the apps/bilig monolith, which serves the built web shell and backend APIs.
• Run WorkPaper and browser performance contracts from packages/benchmarks, scripts/, and e2e/tests.
• Build the AssemblyScript WASM kernel with pnpm wasm:build.

Current Status

bilig is early, serious infrastructure. The architecture is broad and the correctness bar is intentionally high, but it is not a finished Excel clone.

Known open areas include:

• full Excel formula parity
• defined names, tables, structured references, and deeper dynamic-array support
• worker-first browser runtime as the default boot path
• final durable multiplayer sync backend
• typed binary agent frames end to end
• more public package release hardening

Headless WorkPaper In Five Minutes

Start here when you want to use the spreadsheet engine from Codex, Claude Code, a service, or a Node script without opening the browser UI.

@bilig/headless is production-ready for applications that call the documented WorkPaper API directly. The package README is the contract for install, API usage, persistence, validation, supported scope, and agent workflow: packages/headless/README.md. For the shortest public method map, start with the WorkPaper read/write cheat sheet.

Install from npm:

pnpm add @bilig/headless

Try the package without cloning the monorepo:

mkdir bilig-headless-eval
cd bilig-headless-eval
npm init -y
npm pkg set type=module
npm install @bilig/headless
npm install -D tsx typescript @types/node
curl -fsSLo eval.ts https://proompteng.github.io/bilig/npm-eval.ts
npx tsx eval.ts

That maintained TypeScript file builds a formula-backed workbook, edits source data, serializes the document, restores it, and verifies that the recalculated value survives the round trip.

For a runnable external-consumer example, start with examples/headless-workpaper. The repository smoke test executes that same example against packed local runtime packages with pnpm workpaper:smoke:external.

For backend adoption, see docs/node-service-workpaper-recipe.md. It shows a minimal Node service boundary that reads computed summaries, applies one controlled edit, and persists the WorkPaper document.

For tabular service payloads, see docs/csv-shaped-workpaper-input-recipe.md. It normalizes a small CSV-shaped fixture into a WorkPaper workbook and reads formula-backed summary values.

For JSON service/API payloads, the runnable example includes npm run json-records. It maps an array of opportunity records into WorkPaper.buildFromSheets(), adds formula-backed summary cells, and validates exact computed output before printing JSON.

For billing-style service payloads, the runnable example includes npm run invoice-totals. It calculates line-item totals, subtotal, tax, and grand total formulas, then validates exact computed and serialized formula readback.

For reporting and finance automation, the runnable example includes npm run budget-variance. It compares budget and actual rows, calculates dollar and percent variance, and flags rows that need review with a formula-backed alert.

For subscription revenue forecasting, the runnable example includes npm run subscription-mrr. It models starting customers, churn, expansion, and new customers, then prints starting MRR, ending MRR, net expansion MRR, and verified formula readback.

For sales-ops quote workflows, the runnable example includes npm run quote-approval. It calculates list total, discount amount, quote total, max line discount, and an approval flag from formula-backed quote rows.

For operations planning, the runnable example includes npm run fulfillment-capacity. It compares forecast order volume with available labor hours, calculates required hours, capacity gap, short days, and a formula-backed status.

For formula errors, see docs/unsupported-formula-troubleshooting-recipe.md. It shows how to read #VALUE!/#NAME? display text together with structured diagnostics so services and agents can reject or normalize unsupported inputs.

That example also includes npm run agent:verify, a small agent writeback demo that records the exact assumption cells changed, verifies dependent formula readback, persists the workbook, restores it, and proves the formulas and values survived the round trip.

For a tool-calling shape closer to agent SDKs, run npm run agent:tool-call. It returns a compact tool call, before/after computed values, formula contracts, persistence proof, and round-trip verification.

For agent framework wrapper shapes, run npm run agent:framework-adapters. The example keeps the same validated WorkPaper read/write functions and exposes thin framework adapter shapes for AI SDK, LangChain, Mastra, LlamaIndex.TS, LangGraph.js, CopilotKit, and Cloudflare Agents.

For an MCP-style shape, run npm run agent:mcp-tools. It returns a dependency-free tools/list response, a tools/call read, and a verified input edit with structured computed readback. Run npm run agent:mcp-stdio when you want the same tools over newline-delimited JSON-RPC stdio. The package-level stdio binary is bilig-workpaper-mcp, runnable with npm exec --package @bilig/headless -- bilig-workpaper-mcp. It is published in the official MCP Registry as io.github.proompteng/bilig-workpaper: https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.proompteng%2Fbilig-workpaper. For Claude Desktop bundle installs, build build/mcpb/bilig-workpaper.mcpb with pnpm mcpb:workpaper:build; the MCPB guide documents the manifest and verification prompt.

Quickstart:

import {
  WorkPaper,
  createWorkPaperFromDocument,
  exportWorkPaperDocument,
  parseWorkPaperDocument,
  serializeWorkPaperDocument,
  type WorkPaperCellAddress,
} from '@bilig/headless'

type NumericCell = {
  value: number
}

const workbook = WorkPaper.buildFromSheets(
  {
    Revenue: [
      ['Region', 'Customers', 'ARPA', 'Revenue'],
      ['West', 20, 1200, '=B2*C2'],
      ['East', 30, 250, '=B3*C3'],
      ['Central', 18, 300, '=B4*C4'],
    ],
    Summary: [
      ['Metric', 'Value'],
      ['Total revenue', '=SUM(Revenue!D2:D4)'],
    ],
  },
  { maxRows: 1_000, maxColumns: 100, useColumnIndex: true },
)

function numberValue(cell: unknown): number {
  if (typeof cell === 'object' && cell !== null && typeof (cell as NumericCell).value === 'number') {
    return (cell as NumericCell).value
  }
  throw new Error(`Expected numeric cell value, got ${JSON.stringify(cell)}`)
}

const revenue = workbook.getSheetId('Revenue')
const summary = workbook.getSheetId('Summary')
if (revenue === undefined || summary === undefined) {
  throw new Error('Workbook sheets were not created')
}

const at = (row: number, col: number): WorkPaperCellAddress => ({
  sheet: summary,
  row,
  col,
})

const before = numberValue(workbook.getCellValue(at(1, 1)))
workbook.setCellContents({ sheet: revenue, row: 1, col: 1 }, 32)

const saved = serializeWorkPaperDocument(exportWorkPaperDocument(workbook, { includeConfig: true }))
const restored = createWorkPaperFromDocument(parseWorkPaperDocument(saved))
const restoredSummary = restored.getSheetId('Summary')
if (restoredSummary === undefined) {
  throw new Error('Summary sheet was not restored')
}

const after = numberValue(
  restored.getCellValue({
    sheet: restoredSummary,
    row: 1,
    col: 1,
  }),
)

console.log({ before, after, sheets: restored.getSheetNames(), bytes: saved.length })

Rules for agents:

• Use public package exports from @bilig/headless; do not reach into src/ or dist/ unless the task is to change the package itself.
• Addresses are zero-based { sheet, row, col }; resolve sheet ids with getSheetId().
• Use exportWorkPaperDocument() and createWorkPaperFromDocument() for persistence round trips.
• Add tests before changing config rebuilds, range bounds, formulas, persistence, or structural edits.
• Run focused headless tests first, then pnpm publish:runtime:check, pnpm workpaper:bench:competitive:check, and pnpm run ci before publishing or claiming production readiness.

For workflow feedback from people building Node services or agent tools, use the GitHub discussion: https://github.com/proompteng/bilig/discussions/157.

For the five runnable service-workflow examples as one shareable thread, see: https://github.com/proompteng/bilig/discussions/213.

For the search-friendly version with commands and current output snippets, see docs/workbook-automation-examples-node.md.

For the first technical adoption article, see docs/why-agents-need-workbook-apis.md. It explains why agents should operate on workbook APIs instead of spreadsheet screenshots.

For a concrete framework-neutral agent tool loop, see docs/agent-workpaper-tool-calling-recipe.md. It wraps WorkPaper reads, validated writes, computed before/after checks, and persistence into a small tool surface.

For the persistence-focused follow-up article and runnable example, see docs/persisting-formula-backed-workpaper-documents-in-node.md and examples/headless-workpaper/persistence-roundtrip.ts. For service storage, run the typed Postgres JSONB, Redis/KV, and object-storage adapters in examples/serverless-workpaper-api/persistence-adapters.ts.

For the benchmark-focused explainer, see docs/what-workpaper-benchmark-proves.md. It states the 46/46 mean-win claim and the known p95 caveat without turning the benchmark into a blanket performance claim.

For a local benchmark command walkthrough, see docs/local-workpaper-benchmark-walkthrough.md. It shows how to verify the checked-in artifact, run a reduced local smoke benchmark, and compare benchmark diffs.

For a concise HyperFormula comparison and evaluation path, see docs/hyperformula-alternative-headless-workpaper.md.

For a broader headless spreadsheet-engine comparison across @bilig/headless, HyperFormula, IronCalc, ExcelJS, Formula.js, Hucre, Formualizer, and JSpreadsheet Formula Pro, see docs/headless-spreadsheet-engine-comparison.md.

If your search starts with "JavaScript spreadsheet library," use docs/javascript-spreadsheet-library-headless-node.md to separate browser grids, XLSX file tooling, formula engines, and backend WorkPaper runtimes before choosing a package.

For a runnable revenue-model walkthrough, see docs/building-a-revenue-model-with-headless-workpaper.md and examples/headless-workpaper/revenue-scenarios.ts.

For the current Excel-compatibility boundaries, see docs/where-bilig-is-not-excel-compatible-yet.md. It names the macro, formula, XLSX corpus, and UI-claim gaps without treating the project as a complete Excel clone.

For a short guide to interpreting XLSX cached-result corpus reports, see docs/xlsx-corpus-verifier-walkthrough.md.

For a formula-edge fixture walkthrough covering the exact-match XLOOKUP path, see docs/formula-edge-xlookup-exact-fixture.md.

For a formula-edge fixture walkthrough covering paired-criteria SUMIFS, see docs/formula-edge-sumifs-paired-criteria-fixture.md.

For a formula-edge fixture walkthrough covering grouped dynamic-array GROUPBY output, see docs/formula-edge-groupby-spill-fixture.md.

For the published DEV article, see https://dev.to/gregkonush/why-agents-need-workbook-apis-instead-of-spreadsheet-screenshots-3d61. The source mirror with front matter is docs/dev-to-workbook-apis-post.md.

Quickstart

Use Node 24+, Bun, and pnpm@10.32.1.

pnpm install
pnpm wasm:build
pnpm typecheck
pnpm test
pnpm dev

The default dev command runs the local web shell and monolith together.

Useful alternatives:

pnpm dev:web
pnpm dev:web-local
pnpm dev:sync

Local Docker Compose

docker compose up --build

This brings up:

• http://localhost:3000 for the monolith web shell with /v2, /api/zero/v2, and /zero
• http://localhost:4321/healthz for the monolith app runtime
• http://localhost:4848/keepalive for Zero cache
• postgresql://bilig:bilig@localhost:5432/bilig for Postgres

To reset local state:

docker compose down -v

Package Map

Path	Role
apps/web	Vite/React browser source compiled into the monolith
apps/bilig	Fullstack monolith runtime, API surface, and static asset server
packages/protocol	Shared enums, opcodes, constants, and protocol types
packages/formula	A1 addressing, lexer, parser, binder, compiler, JS evaluator
packages/core	Workbook engine, scheduler, snapshots, selectors, sync ownership, WASM facade
packages/headless	Headless WorkPaper runtime surfaces
packages/zero-sync	Zero schema, workbook queries, projection, and event payload helpers
packages/binary-protocol	Wire format for sync frames
packages/agent-api	Agent request, response, event, and framing model
packages/worker-transport	Engine host/client bridge for worker execution
packages/renderer	Custom workbook reconciler and workbook DSL
packages/grid	Reusable React spreadsheet UI components and hooks
packages/wasm-kernel	AssemblyScript/WASM compute fast path
packages/storage-browser	Browser-side persistence
packages/storage-server	Server-side storage integration points
packages/excel-fixtures	Checked-in formula parity fixtures
packages/benchmarks	Benchmark harness and performance contracts

Verification

The repo has a strict local preflight. For small changes, run the narrowest targeted command first; before publishing, use the full gate.

pnpm lint
pnpm typecheck
pnpm test
pnpm test:browser
pnpm bench:smoke
pnpm run ci

Generated sources are checked in and enforced:

pnpm protocol:check
pnpm formula-inventory:check
pnpm workspace-resolution:check
pnpm workpaper:bench:competitive:check

Performance Work

The WorkPaper track is the repo's performance-leadership program. It compares bilig's spreadsheet runtime against HyperFormula-style workloads and keeps the important claims tied to benchmark artifacts, counters, and docs.

<p align="center"> <img src="docs/assets/workpaper-benchmark-card.png" alt="WorkPaper benchmark evidence card with 46 out of 46 comparable mean wins and the visible p95 caveat" /> </p>

Current public evidence:

• packages/benchmarks/baselines/workpaper-vs-hyperformula.json, generated at 2026-05-08T15:00:27.603Z, records WorkPaper 46/46 mean wins on scorecard-eligible comparable workloads: 38/38 public and 8/8 holdout.
• docs/assets/workpaper-benchmark-card.png is the shareable chart for the current scorecard. It is generated from the checked-in artifact with pnpm docs:benchmark-card:generate.
• docs/headless-workpaper-benchmark-evidence.md explains what is measured, what is excluded, and why this is a mean-win claim rather than a blanket p95 guarantee.

Start here:

• docs/workpaper-engine-leadership-program.md
• docs/headless-workpaper-benchmark-evidence.md
• docs/what-workpaper-benchmark-proves.md
• docs/local-workpaper-benchmark-walkthrough.md
• docs/workpaper-oracle-sota-performance-design-2026-04-21.md
• docs/workpaper-oracle-validated-performance-design-2026-04-26.md
• docs/workpaper-oracle-benchmark-expansion-performance-plan-2026-04-28.md

Run the competitive benchmark with:

pnpm bench:workpaper:competitive

Architecture Docs

Good entry points:

• docs/architecture.md
• docs/public-api.md
• docs/formula-language.md
• docs/agent-api.md
• docs/local-first-realtime-loop.md
• docs/binary-protocol.md
• docs/wasm-runtime-contract.md
• docs/testing-and-benchmarks.md

Contributing

Read CONTRIBUTING.md before opening a PR. If this is your first patch, start with the new contributor guide and then claim a scoped starter issue. The highest-value contributions are usually:

• formula parity fixtures and semantic tests
• WorkPaper benchmark scenarios with clear expected behavior
• focused engine correctness fixes
• grid accessibility and keyboard-behavior improvements
• docs that turn existing architecture notes into runnable examples

The shortest public on-ramp is the starter issues queue. Current starter issues are scoped around small runnable examples with explicit acceptance commands, so a first contribution can improve the public WorkPaper evaluation path without understanding the whole engine. If this is your first contribution to bilig, start with the first-timers-only filter.

Please keep changes small, tested, and tied to the package that owns the behavior.

CI

Forgejo Actions is the primary CI surface for this repo via .forgejo/workflows/forgejo-ci.yml. GitHub Actions mirrors the verification contract in .github/workflows/ci.yml.

The strict gate includes frozen lockfile install, full pnpm run ci, artifact budget checks, browser smoke, and tracked-file cleanliness checks.

License

MIT.