Documents with intelligence

Stencila is a platform for creating and publishing dynamic, data-driven documents. Author executable content in Markdown, harness AI agents and workflows, and publish as beautiful, interactive articles.

curl -LsSf https://stencila.io/install.sh | bash

Skills

Reusable instruction sets for AI agents, loaded on demand. Skills follow the open Agent Skills Specification for portable, context-efficient guidance.

Learn about Skills →

Agents

Pair large language models with research-focused tools in a programmable agentic loop. Agents read files, run commands, search the web, and iterate until the task is done.

Learn about Agents →

Workflows

Orchestrate multi-stage AI tasks as directed graphs. Chain agents into pipelines with branching, retries, parallel execution, and human-in-the-loop gates.

Learn about Workflows →

Author in Markdown, Publish Anywhere

Write in Stencila Markdown with executable code, math, citations, and AI instructions — then convert to any output format.

Formats

Import from and export to Markdown, Jupyter Notebooks, LaTeX, JATS, Word, PDF, HTML, and many more. Conversion goes through a canonical schema so any format can talk to any other.

Browse Formats →

Execution Kernels

Run code chunks in Python, R, Node.js, Bash, SQL, and more — directly inside your documents. Outputs render inline alongside your prose.

Browse Kernels →

Themeable & Publishable

Control every visual detail with a token-based design system. Publish to the web, PDF, Word, email, and more — all from the same source document.

Design Tokens

A comprehensive set of CSS custom properties for typography, colors, spacing, plots, and every document node type. Theme once, render everywhere.

Theme Tokens →

Sites

Publish collections of documents as fully searchable, navigable websites with configurable layouts, sidebars, and search.

Site Config →

Multi-Target Output

Render to web, PDF, Word, and email from a single theme. Plot tokens carry across to Plotly, Vega-Lite, and ECharts for consistent data visualization.

Output Targets →

Safe by Design

Tool Guards

Every agent tool call passes through a guard before execution. Guards inspect commands, file paths, and URLs, returning Allow, Warn, or Deny verdicts based on configurable trust levels. Destructive operations like git push --force are blocked even inside sandboxes.

Tool Guards →

Workspace Database

Every document change is tracked in a local SQLite database with full version history, branching, and sync. Audit logs capture all agent tool call verdicts for post-hoc review.

Database Docs →