Engineering, product updates, and behind-the-scenes.
We quietly shipped a round of stability and security improvements to Tripplet's auth system this week. Password reset tokens, rate limiting, and error handling all got tightened up. Here's what we changed and why it matters.
Our three models just got a major upgrade. Taipei 3.1 brings deeper reasoning with extended thinking. Majuli 3.1 cuts response times by 40%. Suzhou 3.1 produces richer creative output across writing, code, and image prompts. Here's what changed and why.
Tripplet's Code workspace lets you describe an app in natural language and get working files in seconds. Under the hood, it's a chat-driven code generation pipeline with live preview, multi-file editing, and one-click deploy. We walk through the architecture decisions.
Extended Thinking lets Tripplet's models spend more time reasoning before responding. It's not always faster, but for complex problems — math, multi-step logic, code architecture — it produces significantly better results. We explain how the mechanism works and when to toggle it on.
We released Tripplet under the Unlicense — anyone can use, modify, and ship it. This wasn't an obvious decision for an AI product. We share our reasoning, the tradeoffs we considered, and how open source has shaped our development process and community.
Not every question needs the most powerful model. We built a routing system that analyzes your prompt's complexity and picks the optimal model — balancing speed and quality in real time. Here's how the classifier works.
When a conversation requires current information, Tripplet can search the web, pull relevant sources, and synthesize the results into its response. We explain the search pipeline, source ranking, and how we present citations.