Open Source

New Free Tool Lets You Edit What AI Is Thinking Mid-Sentence

You've watched AI get things wrong. Now you can stop it mid-sentence.

Deep Dive

A solo developer who goes by onPanda has spent two years building a free tool that cracks open AI models so you can see and touch every small piece of what they produce. Those pieces are called 'tokens' — basically a word or a fragment of a word. Hover over one, click a suggested alternative, or type your own replacement, and the AI keeps writing from your new version. You can edit everything it shows you, including its step-by-step reasoning and the tools it decides to use.

Think of it like editing a document while the AI is still typing, rather than rewriting the whole essay afterward. For anyone building with AI, that saves hours of trial and error. For the rest of us, it's a peek at where these tools are heading: less like a magic black box, more like something you can grab the steering wheel on. Catching a bad assumption at token three is far cheaper than discovering it three paragraphs later.

The tool goes well beyond text. It handles images, video and audio, connects to outside services so the AI can actually do things, and includes a 'browser agent' — software that runs inside your web browser with no setup, clicking links, searching, reading local files and remembering what it learned last time. It also plugs into popular coding assistants like Claude Code and Codex, so you can compare what each one is secretly doing under the hood.

Here's the honest catch: the developer describes it as built for 'geeks, power users and engineers.' This is not a polished app for your parents. Expect a genuine learning curve and a busy screen. And if you care about privacy, you'll need to run it yourself from GitHub rather than using the free web version.

Key Points
  • You can pause an AI mid-answer, swap out a word, and let it keep going — no full restart
  • It also works on AI agents: software that clicks, searches and takes actions on your behalf
  • Free in your browser, or self-hosted from GitHub if you'd rather keep your data private

Why It Matters

More control over AI means fewer embarrassing errors and a clearer view of how it decides things.

📬 Get the top 10 AI stories daily