AI tools · ten tools, twenty guides
AI tools that stay on your device
Ten free tools for working with language models and AI-generated content: check a long chat, pack a PDF, test chunks, plan an agent, read the terms. Each runs in your browser, uploads nothing and is explained in two guides.
Why these tools run in your browser
Most AI tooling asks you to paste your conversations, documents and contracts into someone else's server. These tools do the opposite: the analysis is ordinary JavaScript running in your own tab, so a long chat, a PDF or a contract never leaves your device.
That choice has a trade-off, which every page states: these are transparent counters, patterns and rankings, not AI models. They measure what can be measured — token growth, instruction lapses, chunk boundaries, clause changes, claim wording — and show the reason behind each result, so you can check it instead of trusting a score.
Each tool is paired with two guides that explain the idea behind it with cited sources, and each guide links back to its tool. Start from the problem you have in the table below.
Which tool for which problem
| If you want to… | Use | You get |
|---|---|---|
| A long AI chat stopped following my rules | Context Drift Heatmap | A turn-by-turn heatmap of token growth, instruction lapses and contradictions |
| I want to search PDFs offline and cite the page | PDF to Knowledge Pack | A BM25 index with page citations and one offline HTML file to keep |
| I want to give a chatbot my background without oversharing | LLM Context Passport | A structured passport with personal data flagged and redacted |
| I am choosing a chunk size for retrieval (RAG) | Chunking Strategy Comparator | Seven chunkers compared on your own document, with retrieval results |
| I have a product idea and need an agent plan | Idea to Agent Spec | Tasks, tool schemas, acceptance tests and eval prompts on one screen |
| I want to know which duties in a job AI touches | Job Description AI Impact Analyzer | Each duty marked automated, augmented or human-led, with reasons |
| I want to know how an AI crawler sees my page | AI Crawler HTML Inspector | Main content, heading chunks, summaries and citation anchors |
| The terms of service changed and I want to know what is different | Terms Change Risk Diff | A clause-level diff classified by risk, in plain English |
| I need the claims and the evidence in a research paper | Claim–Evidence Map | Claims, supporting evidence, wording strength and anchors |
| I want controlled variations of an image prompt | Image Prompt DNA | A breakdown by part and variants that change one thing at a time |
All 10 tools, each with two guides
Every tool has its own page with a method section and FAQ, and two guides that explain the ideas behind it with cited sources.
Conversations & context
Keep a chatbot useful over a long session, and decide what to tell it in the first place.
Paste a long AI chat to see, turn by turn, where rules lapsed, answers ballooned and facts changed.
Build a reusable AI context passport and scan it for common personal-data patterns and secrets before pasting.
Documents & retrieval
Turn documents into something searchable, check how they were cut, and map what a paper actually claims.
Turn PDFs into a searchable knowledge base: per-page citations, BM25 search and one offline HTML pack. Runs in the browser.
Compare seven chunking strategies on your own text and questions, with cut answers, retrieval hits and token cost.
Maps a paper's claims, the sentences linked to them, and where strong wording has no number or citation.
Agents & product
Plan an AI agent before you build it: what it does, which tools it may call and how you will know it works.
Turn a product idea into agent tasks, tool schemas, acceptance tests and eval prompts, with a workflow-or-agent check.
Web & content
See your page the way a text-only crawler and a retrieval pipeline would.
See the text a non-rendering fetch gets from pasted HTML, with main content, heading chunks and a rule-based audit.
Work & law
Read a job description or a contract with the reasons visible.
Paste a job description; get each duty labelled Could be automated, Augmented or Human-led, with the reasons.
Compare two terms or privacy policy versions clause by clause, with risk categories and changed numbers.
Creative
Treat an image prompt like an experiment: break it into parts and change one at a time.
Break an AI image prompt into genes, flag gaps and conflicts, then make controlled variations. Runs in your browser.
Questions
Are these tools private?
Yes. Every analysis runs in your browser tab. The pages make no upload requests, need no account and store nothing unless you export a file or tick a clearly labelled save option. You can confirm this in your browser's network panel.
Do the tools use AI models?
No. They use counters, patterns, rankings and word lists you can inspect, such as BM25 keyword ranking, sentence splitting and rule tables. That is why each result shows its reason, and why results are estimates rather than judgements.
Do they work offline?
After a page has loaded once, its analysis does not need the network. Pages you have opened are saved for offline use when your browser allows it; the install page explains the optional offline download.
How accurate are the results?
They are heuristics and estimates, and each page says what it cannot know. Token counts are estimates, exposure labels describe tasks rather than predict jobs, and clause classification is a first pass that does not replace reading the original or getting legal advice.
Why does every tool have two guides?
A tool answers what a text looks like; a guide answers what to do about it. Each guide cites its sources, corrects common misconceptions the sources contradict, and links back to its tool.
Can I use the exported results in my own work?
Yes. Markdown, CSV and JSON exports are yours to use. Keep each page's stated limits in mind when you quote a result, and cite the page you used.