All digital products have an environmental impact — Candli is no exception. We are committed to minimizing our footprint and promoting sustainability in the following ways:
The web is built around JavaScript, a highly dynamic language that is not always energy efficient. Fortunately, the emergence of WebAssembly — a typed binary format for the web — allows developers to deploy much more efficient code. Candli leverages WebAssembly for computationally intensive tasks on the client side, such as image segmentation and tutorial hint generation.
The rest of our codebase is written in TypeScript, a typed superset of JavaScript. This reduces runtime dynamism and allows browsers to apply more aggressive optimizations.
Both our WebAssembly components and our server-side backend are written in Rust, a low-level, statically typed language that enables powerful compile-time optimizations. This ensures that both client and server processes consume as little energy as possible during runtime.
Transferring and storing data consumes energy — less data means less energy used. We apply this principle by:
These optimisations ensure that users enjoy fast loading times while minimizing bandwidth and energy use.
Manufacturing digital devices — especially chips — involves high embodied energy, much of it powered by coal in regions like Asia. To reduce our contribution to this footprint, we focus on maximizing hardware lifespan.
As of June 2025, the desktops used in our office average 6 years of age, significantly longer than the 1.5-year average in the typical 3-year replacement cycle. We also carefully evaluate new web features: before adopting them, we verify that they remain compatible with reasonably old hardware and browsers, striking a balance between innovation and sustainability. This contributes to sustainability because it avoids adding incentive to our users to replace working old hardware.
Because Candli is a web app, data is downloaded whenever the app is used. To limit this:
This minimizes the energy needed to transfer data globally, while improving performance for users.
Modern AI — particularly large language models (LLMs) — offers powerful capabilities but requires vast amounts of energy, especially during inference on large-scale infrastructure.
We use LLMs internally to improve development workflows but do so responsibly:
As we prepare to offer LLM-based tools to Candli users, we commit to using compact models (8B–12B parameters) — a small fraction of the size of frontier models like GPT-4o (~1800B). Additionally, we will raise awareness among users about the environmental costs of using such models, reinforcing mindful usage in educational settings.