What is Jantt?
Jantt turns a JSON document of tasks, dates, dependencies and milestones into an interactive Gantt chart you can drag, link and edit, and writes the changes back to the JSON. The pitch is aimed at language models: instead of asking ChatGPT or Claude to write React or SVG for a timeline, you give it Jantt's schema and cheatsheet, it returns plain JSON, and the engine handles layout, the dependency graph and the critical path.
The core, @jantt/core, is TypeScript with no runtime dependencies and under 14 KB gzipped, per the README. Around it sit packages for React, Vue, Svelte, a web component, a script tag, Python and Jupyter, a CLI (npx jantt open) that syncs edits in the browser back to the file on disk, a Yjs binding for multi-user editing, import and export for Microsoft Project XML and Primavera P6, and @jantt/mcp, an MCP server with six planning tools for Claude Desktop, Cursor and other agents.
It is MIT-licensed and built by one developer, AhmadHassan-BTed, with a live playground on GitHub Pages.
Is Jantt worth it?
Jantt's central argument is right. Language models are poor at layout code and good at structured data, so a schema the model fills in, and a deterministic engine that resolves dependencies and draws the bars, should produce fewer broken timelines than generated JSX. The CLI's two-way sync makes the loop practical: an agent writes project.json, a person drags tasks in the browser, and the file on disk changes for the agent's next turn.
The feature list is long for a project created on 28 August 2026: critical path, earned value management, DCMA-14 schedule audits, Monte Carlo risk runs, Kanban and analytics views, a canvas renderer the README says handles 100,000+ tasks at 60 FPS, and Microsoft Project and Primavera interchange. The README claims models use 10 times fewer tokens than with JSX; it publishes no measurement behind that figure, and the 0-hallucination language around the schema is marketing, not something a schema can promise.
The numbers around the repository do not line up neatly. It shows 6,207 stars but only 17 forks and one contributor, and @jantt/core was downloaded 529 times on npm in the last week, @jantt/mcp 177 times. Far fewer people appear to have built on Jantt than the star count suggests. There is a single tagged release, v1.1.0, from 2 November, and 12 open issues.
For a small team that wants an agent to keep a plan file and a human to adjust it visually, it is worth an afternoon in the playground. For anything that must open cleanly in Microsoft Project or survive a scheduling audit, test the import and export paths against your own files first; Frappe Gantt is simpler and DHTMLX Gantt is far more proven.
Where it earns its place
A developer building an AI project-planning assistant
Show the model's plan as a chart users can correct
Put the Jantt cheatsheet in the system prompt, validate the returned JSON against jantt.schema.json, and render it with the @jantt/react component; onCommit returns the edited plan to send back to the model.
An engineering manager at a 10-person startup
Keep a roadmap file in the repo that both Claude Code and people can edit
Run npx jantt open ./project.json; the agent edits the file, teammates drag tasks in the browser, and the CLI writes changes back to disk. Free, no hosted service.
A data scientist in Jupyter
Turn a pandas table of tasks into a timeline
pip install jantt and call jantt.show(df) with id, label, start, end and dependsOn columns; jantt.streamlit(df) does the same in a Streamlit app.
Before you commit
- Setup
- npm install @jantt/react @jantt/core for React, npx jantt open <file.json> for the CLI, npx @jantt/mcp for the MCP server; Node.js 20 or later.
- Integrations
- React, Vue 3, Svelte, web component, script tag, Python, Jupyter and Streamlit; Yjs for real-time collaboration; Microsoft Project XML, Primavera P6 XER and CSV import and export; MCP for Claude Desktop and Cursor.
- Data and privacy
- Runs entirely client-side or on your machine; plans live in your JSON files or browser local storage. MIT licence.
- Cost at scale
- No licence cost; only the model tokens your agent spends writing the JSON.
- Limits
- Single maintainer; 12 open issues; performance and interchange claims come from the README, not independent tests; the schema is at v1.4.0, so check which version a model was prompted with.
What’s good
- JSON schema and cheatsheet written for LLMs to fill in
- Zero-dependency core under 14 KB gzipped
- CLI syncs browser edits back to the JSON file
- MCP server with six planning tools
- Bindings for React, Vue, Svelte, plain HTML and Python
What’s not
- One contributor and 17 forks behind 6,207 stars
- Token-saving and 0-hallucination claims are unmeasured
- Enterprise features such as EVM and P6 export are young and untested at scale
- Only one tagged release
Best for
Developers building agents or internal tools that generate project plans and want a lightweight, editable Gantt view of the model's JSON.
Jantt: common questions
- Is Jantt worth it?
- Jantt has a sound idea, letting the model write data and the library draw the chart, but a one-person project with this many enterprise claims needs testing on your own plans before anyone relies on it. We rate it 3.4 out of 5.
- How much does Jantt cost?
- Free under the MIT licence, including every @jantt package on npm. No hosted or paid tier.
- What are the downsides of Jantt?
- One contributor and 17 forks behind 6,207 stars; token-saving and 0-hallucination claims are unmeasured; enterprise features such as EVM and P6 export are young and untested at scale; only one tagged release.
- Who is Jantt best for?
- Developers building agents or internal tools that generate project plans and want a lightweight, editable Gantt view of the model's JSON.
- How much does Jantt cost for a team or in production?
- No licence cost; only the model tokens your agent spends writing the JSON.
- Is Jantt safe for company data?
- Runs entirely client-side or on your machine; plans live in your JSON files or browser local storage. MIT licence.
Answers checked Nov 3, 2026.