AISPA: User-Centric System Prompt Auditing
A framework for auditing hidden system prompts across AI products using eight user-centered protection dimensions.
An evidence-first arXiv reading workspace
Read the paper. Keep the evidence.
Triage abstracts, inspect full-paper claims, ask in separate research threads, explain or rewrite any selection, and carry the useful parts into your own library.
Built for the papers you already read on arXiv
Recent AI papers
Standardized Evaluation for Computer-Use Reward Models
One reading loop
Triage. Investigate. Remember.
Abstract signal, full-paper depth
Quick Take filters the list from the abstract. Deep Take reads available paper sections and ties findings, methods, and limitations back to evidence.
One paper, many research threads
Ask grounded questions, start a fresh conversation when the topic changes, and return to named conversation history later.
A selection becomes a toolbelt
Explain a sentence, summarize a passage, inspect a formula or figure, paraphrase academic prose, or discover papers connected to the selection.
A library you can leave with
Save papers, assign reading status, tags and collections, import Zotero exports, then download clean Markdown or open it in Obsidian.
The current extension, not a roadmap
A complete loop for serious reading.
Quick Take + Deep Take
Move from abstract triage to evidence-linked full-paper analysis without losing the arXiv source.
Grounded Ask sessions
Markdown answers, paper excerpts, separate conversations, rename, revisit, and delete controls.
Explain history
Sentence explanations and every selection workflow are saved per paper, ready to reopen on another visit.
Selection research tools
Summarize, analyze visual context, paraphrase in three modes, or search arXiv from a selected idea.
Research library
Reading states, tags, collections, related papers, Zotero import, notes, and account sync in one side panel.
Your model, your language
Use the private built-in provider or an encrypted personal OpenAI-compatible endpoint. UI and AI replies support 20 languages.
No new research ritual
It meets you on arXiv.
Open a recent, new, search, abstract, or HTML paper page.
Start with Quick Take, then open Deep Take, Ask, or a selection tool only when the paper earns more attention.
Return to conversation and explanation history, save the paper, organize it, or export the result to your own files.
Private by architecture
Credentials stay off the extension.
Built-in model credentials are used only by the Cloudflare Worker and are never returned to the browser. Personal provider keys are encrypted server-side. Markdown exports download directly to your device; temporary PDF storage is used only when processing requires it.
- 01
System API keys never ship in the Chrome package
- 02
Personal provider settings are encrypted at rest
- 03
Research exports remain ordinary local Markdown files
Simple beta pricing
Start free. Upgrade when it earns its place.
Free
$0
- ✓ 5 Quick Takes per day
- ✓ 30 paper questions per month
- ✓ 50 Explain and selection actions per month
- ✓ 30 saved papers
- ✓ Local Markdown export
Pro
$9.90 USD per month
- ✓ 60 Quick Takes per day
- ✓ 1,000 paper questions per month
- ✓ 1,500 Explain and selection actions per month
- ✓ 2,000 saved papers
- ✓ Full export and account sync
Secure checkout is provided by the active billing partner. You will see the provider before leaving this site.