---
updated: 2026-10-16
scope: /
parent: https://sema-lang.com/ai.txt
---

# Sema playground context

https://sema.run is a browser editor, evaluator, debugger, examples library, and
virtual filesystem for Sema. Programs execute locally in WebAssembly. Running
a program does not send its source to a Sema evaluation server. Programs can
make network requests through the browser.

## Browser capabilities and limits

- File operations access the playground's virtual filesystem, not the host OS.
  The default memory backend loses files on reload. Optional localStorage and
  IndexedDB backends persist across reloads; sessionStorage lasts for the tab
  session. Editor contents and preferences are stored separately in localStorage.
- HTTP operations use browser fetch and follow CORS restrictions. A URL that
  works in native Sema may fail in the browser. There is no automatic CORS proxy.
- Shell and subprocess execution are unavailable. Native LLM provider builtins
  such as `llm/complete` are not registered in this playground. Use native Sema
  for those integrations. Sema Web's DOM bindings are a separate package.
- Execution normally uses a worker when cross-origin isolation is available.
  Cancellation is available for worker evaluation. A main-thread fallback exists.

## Use examples and agent tools

Open [Hello World](https://sema.run/?example=getting-started/hello.sema) or
[HTTP basics](https://sema.run/?example=http/fetch-basics.sema).
The `example` query parameter selects a bundled example and replaces restored
editor content; it does not execute the program.

When `document.modelContext.registerTool` is available, the page registers
experimental WebMCP tools. Prefer those tools for editor, evaluator, debugger,
example, and virtual-file operations. Browsers without the API retain the UI.

Start with `read_editor` or `find_examples`. Use `write_editor` or `load_example`
to choose code, then `format_editor`, `run_editor`, and `read_output`. Loading
or writing source does not execute it. Read existing user source before replacing
it. Inspect errors and returned debugger status before reporting success.

The [playground agent guide](https://sema.run/llms.txt) describes the tool
workflow. The [AI Catalog](https://sema-lang.com/.well-known/ai-catalog.json)
links the browser tools to the separate native MCP server.

## Related contexts

- [Sema website](https://sema-lang.com/ai.txt): project and runtime overview.
- [Sema documentation](https://sema-lang.com/docs/ai.txt): code-generation
  guidance and links to individual reference pages.