Lucid

Show HN: I built a web tool to see and edit what an AI thinks before it answers

Last checked:

Visit Lucid

What is Lucid?

Lucid is a browser-based tool that visualizes language model internals using the Jacobian lens, allowing users to see which concepts a model holds at each layer before generating output. It requires no signup and uses Zener cards as visual probes to inspect model representations in real time.

Lucid pricing

Pricing model: Freemium

Free with rate limits (6 questions/minute using provided Claude Sonnet key); unlimited access with personal Anthropic API key

Lucid pros

  • Real-time layer-by-layer visualization of model thinking and internal representations
  • No account or installation required—works directly in browser
  • Sessions are shareable and exportable for collaboration
  • Novel Jacobian lens approach grounded in cognitive science (J-space concept from Anthropic)

Lucid cons

  • Limited to small open models (Qwen 0.5B–3B, Pythia 1.4B) on public instance
  • Rate limited to 6 questions per minute on free tier
  • Requires personal Anthropic API key to access full model selection and remove limits

Frequently asked questions about Lucid

What is the Jacobian lens?

A mathematical approach to reading language model internal representations, based on Anthropic's J-space concept—the workspace where a model holds concepts it can report, redirect, and reason with.

Do I need to sign up?

No. Lucid is accessible directly in your browser with no account or installation required.

Which models can I inspect?

The public instance supports Qwen (0.5B–3B) and Pythia (1.4B). Paste your own Anthropic API key to choose from Claude models (Sonnet 5, Opus 4.8, Haiku 4.5, Fable 5).

Why Zener cards?

They serve as standardized visual stimuli to probe what the model thinks at each layer, inspired by the original Zener cards used in parapsychology experiments for testing mind-reading.

Categories

Use cases

Browse all AI tools on NeedAnAI