DEV Community: Astro

We put a free, unlimited LLM inside a static web page — Transformers.js + Gemma 3, zero servers

Astro — Fri, 30 Oct 2026 05:09:54 +0000

Every LLM demo I have ever seen has the same fine print: bring your own API key, or the demo is rate-limited, or your prompts go to our server. So I built the opposite demo.

DevKit's AI Playground is a chat with a real language model that runs 100% in your browser. No backend, no API key, no cost per token, and — because there is no server at all — nothing you type can leave the page. It's a static HTML file on GitHub Pages, like the rest of DevKit.

How it works

Transformers.js v3 does the inference, inside a Web Worker so generation never freezes the UI.
The model is Google's Gemma 3 270M IT, 4-bit quantized (q4f16) — a ~293 MB one-time download, cached by your browser for future visits. If Gemma can't start on your device (it needs real memory headroom), the page automatically falls back to the smaller SmolLM-135M (~121 MB) in a fresh worker, so the chat still works.
It tries WebGPU first (your graphics chip) and falls back to WASM on the CPU.
Tokens stream in as they're generated, with a live tokens/sec readout — which is either a fun toy or a benchmark, depending on how serious your hardware is.

Click "Load the model", wait for the one-time download, and chat. That's the whole signup flow.

Honest limits

The models are small — 270M and 135M parameters — and live in your RAM. It writes plausible text, answers simple questions, does little creative tasks — and cheerfully makes things up. It is not GPT-class, and the page says so up front. The point isn't to replace an API; it's to show how far "the browser is the runtime" has come, and to give the LLM-curious a playground with literally zero setup and zero bill.

DevKit's other LLM tools are the grown-up counterparts: a cost calculator that prices 14 real API models with cached-read/cache-write splits, and a tokens/sec simulator. The playground is what those calculators look like when the model is small enough to be free.

Also new in this release

Four more client-side tools joined the kit: a JWT decoder (decode-only, and it says so — the signature is not verified), a cron expression explainer (plain English + next 5 run times), a color palette generator, and a Pomodoro timer whose countdown runs in the tab title. Same rules as everything on DevKit: no uploads, no accounts, no tracking, and every calculator's math is checked by automated tests (310 checks and counting).

Try it and tell me the tokens/sec number your machine gets — and which model you'd want to see in the playground next. SmolLM2-360M? Qwen? The download budget is the boss fight.

DevKit guides: every LLM harness and code editor, judged — good, bad, and best for

Astro — Mon, 26 Oct 2026 05:45:06 +0000

I kept picking developer tools by reading scattered docs pages and pricing pages, then finding out the catch later. So I put the comparison in one place instead.

DevKit now has a guides section covering 57 developer tools in 6 categories:

LLM harnesses — 16 tools
Code editors — 11 tools
Terminals — 8 tools
API clients — 8 tools
Git clients — 7 tools
Database tools — 7 tools

Each entry gives the same treatment: a verdict, what it is, what it is good at, what it is bad at, who it is best for, and its price, with links to the official site and docs. The guides were checked in Oct 2026, and the links were verified before publishing.

LLM harnesses

The 16 LLM harnesses split into frameworks, agent frameworks, gateways and routers, and local runners. A few examples from the guide:

LangChain — the Swiss-army framework. Largest integration ecosystem of any LLM framework, but the API has changed often and older tutorials go stale fast. Best for teams building varied LLM apps who want the biggest ecosystem and don't mind the learning curve.
LlamaIndex — the RAG specialist. Best-in-class document ingestion, parsing and retrieval. Best for teams whose core problem is RAG over documents.
LiteLLM — one OpenAI-compatible API for 100+ providers, self-hosted. Best for teams that want provider freedom and cost control without vendor lock-in, and can run a proxy.
Ollama — the easy button for local models. One command and you're chatting offline, with models staying on your machine. Best for developers trying local models or prototyping offline.

Code editors

The 11 code editors split into graphical editors and terminal editors:

VS Code — the default choice for most developers. Free, huge extension ecosystem, good at everything. Best for almost anyone.
Cursor — the leading AI-first editor, a VS Code fork where the agent, not the sidebar, is the point. Cursor restructured its pricing in Oct 2026: a free tier with limited agent use, then Plus $8/mo, Premium $30/mo, Super $100/mo, and Ultra $200/mo. Heavy agent use burns through the included usage pool fast.
Devin Desktop (formerly Windsurf) — a strong AI agent editor, rebranded as Devin Desktop after Cognition acquired Windsurf. Its Pro plan rose from $15 to $20/mo in March 2026. Best for developers who want an agentic editor with a good free tier.
Zed — the fastest graphical editor. Native Rust speed with real-time collaboration built in, but a far smaller extension ecosystem than VS Code's.

A few things that changed while I was writing these

Warp open-sourced its terminal client under AGPL-3.0 in April 2026. The AI and cloud platform stays proprietary, and the terminal still asks for an account for AI and team features.
Cursor restructured its pricing in Oct 2026, as above — check the current tiers before you budget for heavy agent use.
Windsurf is now Devin Desktop, after Cognition acquired it. The editor is still a VS Code fork, but the rebrand makes its long-term roadmap less certain than Cursor's.

Try it

Start at the DevKit guides hub, pick a category, and read the verdicts for the tools you are choosing between. The main site is DevKit — 69 calculators, converters and dev tools, all running client-side in your browser, including the LLM cost calculator.

Which tool did the guides get wrong, or which one should be added next? Let me know in the comments — I'd love to hear what you'd actually pick.

DevKit: a one-file calculator library with tested formulas — now with an LLM cost calculator

Astro — Sun, 25 Oct 2026 21:53:41 +0000

I kept reaching for a quick calculator and landing on pages crowded with ads, pop-ups, and distractions. So I built the tool I wanted instead: a lean calculator site that gets out of the way.

DevKit is a one-file calculator library with 69 tools, all running client-side in your browser. The tools are organised into tabbed pages, so you can find the one you need without scrolling through a long list.

What's new: an LLM cost calculator

The newest addition is for anyone working with language models. The LLM cost calculator helps you estimate what a request will cost, and the tokens/sec simulator shows how generation speed affects timing. It's the part I reach for most when I'm planning an LLM feature and want a quick sense of cost and speed before writing code.

Verified formulas

A calculator is only useful if you can trust its answers, so the formulas in DevKit are verified with tests. That gives me a way to catch mistakes when a formula changes, instead of hoping the numbers look right.

Try it

Try DevKit here

What calculator should I add next? Let me know in the comments — I'd love to hear what you'd actually use.