A product from Ripple Design

We built our own AI chat.
You’re invited.

Free, browser-based, top open-source models routed across Cloudflare, Groq, and Cerebras. Open the URL, start typing. Nothing is saved.

Start chatting

No signup. No app. No data retained.

Used by Ripple Design every day for drafting, debugging, and research. Now open to anyone.

Open a URL. Start typing.

No signup. No email field. No "verify your account." You land on the chat, pick a model, ask anything. Close the tab and the conversation is gone. The privacy default that took ChatGPT two years to add is the only one we ship.

Headline 'A URL. That's the whole onboarding.' Browser bar shows ai.rippledesign.co with an arrow into a chat composer with the placeholder 'Ask anything.' Strikethrough list shows what other AI chat tools require (create an account, verify your email, accept the cookie banner, pick a plan, link a payment method), all skipped.

Auto routing across three vendors.

Cloudflare, Groq, and Cerebras all serve the same open-source models at different speeds and price points. The router measures live time-to-first-token, picks the snappiest healthy vendor, and falls back automatically when one slows down. Fast, Balanced, Smart, you pick the bias.

Headline 'Three vendors, one chat. The router decides.' Diagram shows the chat box at the top, three vendor logos below (Cloudflare, Groq, Cerebras) with live latency bars, and an arrow from the fastest one back to the chat. A small segmented control reads 'Fast · Balanced · Smart'.

Search the web, with citations.

Flip the Search toggle and the chat stops guessing. A planner LLM breaks your question into web searches, runs them in parallel, and a synthesizer writes the answer with inline citations to the sources it used. Good for anything that depends on current information.

Headline 'When the answer needs current sources, turn on Search.' Chat composer with a Search toggle highlighted, then an answer streaming below with numbered citation chips that link to source articles in a side panel.

Top open-source models, no API key required.

Llama 3.3 70B and Qwen 3 235B as the workhorses. We add and retire models as the open-source frontier shifts. You pick from the dropdown, we handle the hosting, billing, and fallback chain. The same models you would have to wire up vendor accounts and API keys for, available in your browser.

Headline 'Llama 3.3 70B. Qwen 3 235B. No API key needed.' Model picker dropdown open with multiple open-source models listed, each labelled with a vendor tag, with a checkmark on the selected one.

Why does a design studio build an AI chat tool?

Honest answer: we wanted a place to use open-source models without a third-party SaaS sitting between our team and every prompt we type. The default AI tools save your conversation history for training, billing, or "personalization." We wanted the opposite default.

So we built Ripple AI. It runs on infrastructure we control in Mumbai, routes across three open-source-model vendors based on live latency, and saves nothing for public users. We use it every day for drafting, summarizing, debugging, and quick research.

The deeper reason: Ripple Design is a software studio, not a design shop. We ship products end-to-end. GST Invoice Generator was the first we opened to the public, Ripple Meet was the second, Ripple AI is the third. The door opens on the ones useful enough to share.

Today, and what we’re building next

Use it today for

  • Drafting emails, briefs, and proposals
  • Summarizing long docs, transcripts, and threads
  • Debugging code with a second pair of eyes
  • Plain-English explainers for things you half-understand
  • Quick web research with cited sources
  • Brainstorming when a blank page is the blocker

Building next

  • File uploads for context (PDFs, docs, transcripts)
  • Image inputs for visual questions
  • Deeper web research (full-page fetch, not just snippets)
  • Saved conversations for signed-in admins
  • More open-source models as they release

The one thing not on the roadmap: replacing the frontier closed-source models for the hardest reasoning tasks. For those, GPT and Claude keep the title.

Ready when you are.

Start chatting

Free. No signup. Nothing saved.

How it works

  1. 1

    Click Start chatting

    Opens ai.rippledesign.co in a new tab. No signup screen, no email field, no install.

  2. 2

    Pick a model, or leave it on Auto

    Auto routes across Cloudflare, Groq, and Cerebras based on live latency. Pick Fast for snappy replies, Smart for the heaviest reasoning, Balanced when you do not care.

  3. 3

    Type your question, hit send

    The answer streams in. Toggle "Search" for current info from the web with citations.

Questions

Is it really free? What is the catch?

No catch. We pay for the inference. We use Ripple AI ourselves every day for drafting, debugging, summarizing, and research, and we left the door open so anyone can use it. There is a per-IP rate limit and per-vendor daily caps so a single user cannot burn the whole budget, but normal use never hits them.

What models do you offer?

Llama 3.3 70B and Qwen 3 235B are the workhorses. We add and retire models as the open-source frontier shifts, so the picker is the source of truth. All of them route through Cloudflare Workers AI, Groq, or Cerebras depending on availability and latency.

What is Auto routing?

We measure time-to-first-token across vendors in real time. Fast biases toward the snappiest vendor for short tasks. Smart biases toward the heaviest reasoning model. Balanced splits the difference. If a vendor goes down or gets slow, the router falls back to the next healthy one without you noticing.

What does Search mode do?

Turn on the Search toggle in the composer and the system runs a research fan-out: a planner LLM breaks your question into search queries, runs them in parallel against the web, and a synthesizer model writes the answer with inline citations to the sources it used. Good for anything that depends on current information.

Do you save my chats?

No. Public chats run through a stateless session. Nothing is written to a database. When you close the tab, the conversation is gone. We see aggregate token counts and latency per vendor for health monitoring, never the content of your messages.

Is it as good as ChatGPT or Claude?

For most everyday tasks, you will not notice a difference. The frontier closed-source models (GPT and Claude) still have an edge on the hardest reasoning, the longest context, and the most agentic workflows. For drafting, summarizing, code questions, plain-English explainers, and quick research with citations, the open-source models we route to are entirely competent.

Why did Ripple build an AI chat tool?

Two reasons. One, we wanted a place to run open-source models on infrastructure we control, without a third-party SaaS reading every prompt our team writes. Two, we ship products for a living, and shipping our own products in public is how we prove that. Ripple AI sits next to Ripple Meet and GST Invoice Generator on /tools as evidence we build, not just consult.

Are there usage limits?

A per-IP rate limit on chat requests to keep abuse out, and a per-vendor daily token budget so one bad day with a runaway script does not burn the whole month. Normal interactive use never trips either. If you hit a limit you will get a clear message, not a silent failure.

What is coming next?

On the build queue: file uploads for context, longer running research with deeper fetches, image inputs, and a saved-conversation mode for signed-in admins. The privacy default for the public chat (nothing saved) stays.

Building something for your business? Let’s talk. No pitch, just a conversation.

Book a CallWhatsApp