Our own project: software we built and operate ourselves.
TinyServe
A real language model computes answers live in your browser, via WebGPU. On screen you watch the conversation's memory grow (the KV cache grid), which words the model looks at (attention heatmaps), and how a small model prompts the large one (speculative decoding). What usually happens out of sight on a server, you watch here with your own eyes.
An LLM is a black box, even for engineers
Teams shipping AI often ask: why does a long prompt compute fast while the answer comes out slowly? Why does graphics card memory run out on a longer conversation? Slide decks and blog-post diagrams do not answer that. TinyServe shows these mechanics live, on a real model, with no server and nothing to install.
Four inference mechanics, visible live
The two inference phases rendered separately: why a long prompt computes fast (prefill, parallel) while generating text is slower (decode, one token at a time).
A 3D grid (layers x heads x positions) that fills in live as tokens are generated: you see exactly how much GPU memory a longer conversation actually costs.
A per-layer, per-head attention heatmap: which earlier tokens the model is actually looking at while generating the next one.
A small model proposes a token, the large model accepts or rejects it (speedup with no quality loss); separately, a simulation of a block-based memory allocator (like vLLM's) versus contiguous allocation, with an out-of-memory visualization.
A real model, not an animation for show
models you switch live (Qwen3 0.6B and 1.7B, Llama 3.2, LFM2.5 1.2B, DeepSeek-R1) and compare their behavior with your own eyes.
is the smallest model (Qwen3 0.6B): you run a real language model in the browser with no long wait.
server requests during generation: everything runs on your machine, so nothing leaks out.
Every choice has a reason
Got a nagging problem? Describe it, I'll code the rest.
One sentence is enough to start. Tell me what's bothering you. Within 48 hours I come back with a clear answer: can it be done, and at what fixed price.