Packages

Thirty packages, each one plain Lumen with tests, imported straight from a URL: no package manager, no install step, no lockfile. Those marked FFI link a C library and build on the native target (some also ship a wasm archive for the browser).

Formats 6

Parse and write text formats.

csv CSV parser/stringifier with quoted fields, CRLF, row helpers, and flat arrays. toml Practical TOML parser/stringifier for sections, dotted keys, typed values, and simple arrays. dotenv .env parser for key/value config strings. semver Semantic Versioning parser, comparator, incrementer, and range evaluator. markdown Pure-Lumen Markdown -> HTML renderer (no FFI); render(md). pdf FFI PDF text and metadata via Poppler: layout mode, page ranges, document info.

Data 3

Databases and stores, over the C FFI.

plume FFI Typed data-mapper and migration tool over PostgreSQL, SQLite and MySQL via FFI: explicit field/column mapping, projections, relations, checksummed migrations, transactions. sqlite FFI Typed SQLite access via FFI; open/exec/query int and text over a C shim. pgvector FFI Persistent vector store on PostgreSQL + pgvector via FFI: upsert, cosine search, distance bounds.

Servers 7

HTTP servers, sockets, templates and validation.

rest Router and HTTP server: decorated controllers, a route table checked at startup, JSON replies. openapi OpenAPI 3.0 document generation for rest controllers: @openapi/@schema decorators, cross-referenced against the real Route[]. websocket RFC 6455 websockets: handshake, frames, server and client — a browser connects, exchanges and closes clean. socketio A socket.io server the official browser client connects to: Engine.IO and Socket.IO framing over websocket. sse Server-sent events: one-way push over an HTTP response that never ends, with resumption. press Templates for pages: EJS-style tags, escaped by default. validation Rules written on the fields they govern; faults come back as a list.

AI & agents 3

Model APIs, agents and traces.

ai Typed AI helpers for OpenAI-compatible chat APIs: RAG, conversation memory with context compression, structured output, a live tool-calling agent loop, and an MCP client (HTTP, stdio, SSE). agents Agents, prompts, models, MCP links and sub-agents as database rows — run, evaluated and reconfigured over HTTP without a restart. tracing OpenTelemetry traces for AI runs over OTLP/HTTP: spans, generations, tools, token usage — six backends, Langfuse-compatible.

Tooling 7

Command lines, terminals, schedules and mail.

token-gate Token-optimizing command proxy (tkg): per-command handlers for git/ls/find/du/grep/tables/linters/tests, savings ledger, Claude Code hook. code-index Codebase symbol index (cidx): map/find/refs/search/rank, persistent cache, 8 languages — AI navigation, no vector DB. args Command-line flag and option parsing (hasFlag, optionValue, positional). quickjs FFI Embed a QuickJS JavaScript sandbox via FFI; set/eval/get values across the boundary. cron FFI Schedule expressions and the next time one fires. tty FFI Raw-mode terminal input, key decoding, terminal size, and ANSI cursor/screen control via a C shim (community). mail Sending mail: one Mailer interface, Resend and SendGrid implementations, one HTML shell.

Small & example 4

Tiny packages that show the import model.

mathx Integer math helpers: gcd, lcm, ipow, isPrime, clampInt. geo 2D point helpers: distanceSq, manhattan. hello Minimal greet function; canonical URL-import example. greeter Two-file package; demonstrates recursive URL-relative imports.

Using a package

Import an export straight from the package's URL and build as usual. The source is fetched over HTTPS at compile time and compiled in.

import greet from "https://lumen-lang.org/package/std-contrib/hello/hello.ts";

console.log(greet("world"));   // Hello, world!

lumen-lang.org/package/std-contrib/<name>/<name>.ts redirects to the package's repository, and the raw GitHub URL works too. Each package's page has its API, the decisions inside it and how it is tested; the FFI packages document their local build there.

markdown is fast

The pure-Lumen markdown renderer outpaces the popular Node.js Markdown libraries. Rendering a typical ~3 KB document in a tight loop — renders per second, higher is better:

Renderer	renders / sec
markdown (this package)	6,775
markdown-it	2,020
marked	1,064

About 3.4× faster than markdown-it and 6.4× faster than marked on the same document.

Contributing

One package per directory, statically typed, tested, current-language features only, with an entry in the catalog. The community page has the short version and links to CONTRIBUTING.