Built for the curious
One workspace for chat, code, and research — four specialized models, one conversation. Ask anything to begin.
Free to start · No account required · Your message carries into the chat
One workspace for chat, code, and research — four specialized models, one conversation. Ask anything to begin.
Free to start · No account required · Your message carries into the chat
The lineup
Every model in Tripplet is tuned for a different kind of thinking. Switch between them mid-conversation — your context comes along.
Our most capable model — deep reasoning, agentic tool use, and full code pipelines through Astro 5 Code.
Structured thinking for research, analysis, and multi-step problems that deserve more than a first guess.
The balanced daily driver — fast enough to keep up, thoughtful enough to be worth asking.
Light and immediate. Free for guests, no account required — start a conversation in one click.
Capabilities
Not a chat box with extras bolted on — search, memory, execution, and your own tools, woven into every conversation.
Responses arrive token by token, so you read while the model writes — no spinner, no waiting for the whole answer.
Fresh results pulled into context as you ask, so answers reflect the world as it is today.
Tripplet remembers what matters across conversations. Pick up where you left off, weeks later.
Code runs in a sandboxed interpreter — actual output, actual errors, not a plausible imitation.
Link GitHub, Gmail, Notion, and more. Your assistant works with your tools, scoped to your account.
A built-in knowledge base you can read, extend, and pull into any conversation.
Modes & tones
Modes change the model's approach; tones change its voice. Stack them however the moment calls for.
think
Slows the model down for careful, low-temperature reasoning when precision matters.
deep-research
A longer leash and a bigger budget — thorough, multi-angle answers for hard questions.
web-search
Live results woven into the conversation, cited and current.
study
Explains instead of solves — guides you to the answer so it sticks.
Real execution
When Tripplet writes Python, it executes it in a sandboxed interpreter and shows you the real output — and real errors. No confidently wrong arithmetic, no imagined results.
Want more than Python? Enable the sandboxed Linux skill in Settings and the assistant gets a full in-browser VM with a real shell.
>>> primes = [n for n in range(2, 50)
... if all(n % d for d in range(2, n))]
>>> sum(primes)
328Connectors
Link the apps you already live in, and Tripplet can read and act on them — issues, mail, docs — scoped to your account, revocable any time.
For developers
Every response is server-sent events from the first token. Build on the same streaming API the Tripplet interface runs on — plus an MCP server and OAuth for your own agents.
const res = await fetch('/api/chat', {
method: 'POST',
body: JSON.stringify({
model: 'astro-5',
messages: [{ role: 'user', content: 'Explain quines.' }],
}),
});
// Server-sent events — tokens as they're generated
for await (const chunk of readSSE(res.body)) {
render(chunk.delta);
}Triplepedia
Articles you can read, extend, and pull straight into a conversation — Tripplet's collective memory, one tab away.
Free to start — no card, no setup. Guests get Suzhou 4 instantly; an account unlocks the full lineup.