Company

Mission #

Make the coding agent a tool you own, the way developers own their editor. elelem aims to be the Vim of coding agents: small enough to understand, open for extension, and able to run on your own hardware.

What exists today #

Works today #

• A terminal coding agent. The elelem gem, with a core of 978 lines of Ruby, MIT licensed and published on RubyGems.
• Providers. Claude through the Anthropic API, Ollama, and GGUF models loaded into the elelem process with llama.cpp: no server, no network call.
• Tools and integrations. File, shell, search, and edit tools, MCP servers, and agent skills.
• 9 plugins published on RubyGems, each documented on this site.

In progress #

• OpenAI and Vertex AI providers have known bugs (#26, #24).
• A web UI that is for local use only until it has authentication (#1).
• An eval harness, a ratatui terminal UI, and a lab of experiments in inference: ggml, llama2.c, candle, and stable-diffusion.cpp.

In the open #

Every repository is public at src.mokhan.ca/elelem, and every known bug is tracked publicly at github.com/elelem-sh/elelem.sh/issues. Patches are welcome by email.

Why now #

Coding agents went from novelty to daily tool in a few years. Most of them are large, closed, or tied to one model vendor. Meanwhile, open models keep getting better, and laptops can run them. That opens room for an agent that is model-agnostic, local-first, and simple enough for a team to audit.

Who builds it #

elelem is built by mo khan, a software engineer. It started as a personal harness and grew into a plugin system as the pieces proved useful on their own.

Work with us #

We are interested in hearing from:

• Teams that want a coding agent they can run on their own hardware or audit end to end.
• Model and infrastructure providers who want first-class support in elelem.
• Investors and partners who share the view that developer tools should be small, open, and owned by their users.

Get in touch through mokhan.ca.