Private beta · Chrome extension · Not yet on the Web Store
>_
Intelligence Engine Get early access
Works with Claude · ChatGPT · Perplexity · Gemini

You generate
intelligence
every day.
Keep it.

Intelligence Engine runs quietly while you use AI tools. It captures how you think — the reasoning, frameworks, and approaches that work — anonymizes every word, and stores it locally. Your device only. No server. No account.

Join the private beta See how it works →
Zero servers No account required Free to start
Intelligence Engine
● Capturing
147
Entries
4
Platforms
0
Servers
Recently captured
CLAUDE 2 min ago
Reframe roadmap conversations around constraints, not priorities — once alignment exists on constraints, resource questions resolve themselves.
CHATGPT 1h ago
When [company_a1b2] objected to pricing, the unlock was ROI framing tied to their specific metric, not feature comparison.
PERPLEXITY 3h ago
The "boring stack" principle: default to the established solution unless there's a clear, quantified reason to deviate.
⌘K approach that worked with difficult client
31×Same problems re-researched by heavy AI users. Six hours a week.
−IQFrequent AI use measurably reduces critical thinking (Gerlich, 2025)
76%AI-using professionals pay for at least one AI tool. The buyer is primed.
85%Chrome WebGPU coverage — enough for full on-device Llama-3.2-3B inference.
0 bytesUser data transmitted to any server during normal operation.
31×Same problems re-researched by heavy AI users. Six hours a week.
−IQFrequent AI use measurably reduces critical thinking (Gerlich, 2025)
76%AI-using professionals pay for at least one AI tool. The buyer is primed.
85%Chrome WebGPU coverage — enough for full on-device Llama-3.2-3B inference.
0 bytesUser data transmitted to any server during normal operation.
The problem

You've solved this before.
You'll solve it again.

AI tools help you think. They don't remember how. Every session starts from zero — your hard-won frameworks, the approach that worked, the decision rationale — gone when the tab closes.

You re-research the same problems
One user tracked this for two weeks. They re-researched the same issues 31 times. Six hours a week, thinking they were making progress.
Context doesn't carry across AI tools
You explain the same project from scratch to every model. Your context, your constraints, your vocabulary — rebuilt every time.
Your best thinking evaporates
The framework that cracked the hard problem on Tuesday. The reframe that unlocked the client. Gone. You'll approximate it again next time.
You're generating intelligence every day. None of it compounds.
Intelligence Engine changes that. It captures the signal — the reasoning, not the chat — and gives it back to you, searchable, private, permanent.
How it works

Five steps. On your device.
Nothing leaves.

The pipeline is small by design. Every step runs inside your browser — no network calls after the one-time model download.

01 · OBSERVE
Watches conversations

A lightweight content script observes DOM turns on Claude, ChatGPT, Perplexity, and Gemini as you use them naturally. No keystroke logging. No click tracking. No screenshots.

content_script · MV3 · zero DOM mutation
02 · CLASSIFY
Keeps the reasoning

An on-device LLM (Llama-3.2-3B via WebGPU) reads each segment and decides: does this carry reasoning intelligence? Casual chat, factual Q&A, and task execution are dropped. Only frameworks, decisions, and approaches survive.

WebGPU · Llama-3.2-3B · WASM fallback
03 · ANONYMIZE
Removes identity

BERT-NER identifies every person, company, location, dollar amount, and date. Each entity is rewritten to a deterministic placeholder before storage. If any identifier survives verbatim after two passes — the segment is dropped, not stored.

BERT-base-NER · SHA-256 deterministic replacer · drop-not-store
04 · EMBED
Indexes the meaning

all-MiniLM-L6-v2 converts each anonymized segment into a 384-dimensional semantic vector. Everything lands in your browser's IndexedDB. No cloud database. No sync. It lives exactly where you are.

all-MiniLM-L6-v2 · 384-dim · IndexedDB
05 · RETRIEVE
Semantic search

Ask natural questions — "the approach that worked with a difficult client," "why we chose the boring stack," "how I framed the roadmap conversation." The engine finds the entries you meant, not just keyword matches. Export to any AI tool with one click.

cosine similarity · ⌘K anywhere · export to markdown/JSON/HTML
Anonymization, in one glance

Patterns stay. Identity doesn't.

Every persisted segment goes through this rewrite. Any entity that survives verbatim after both passes is dropped — never stored.

Before · what was said The approach that worked with Sarah Chen at Acme Corp was to reframe the roadmap conversation around the three constraints. Once we had alignment, the $450K budget question in Q3 resolved itself.
→
After · what is stored The approach that worked with [person_a1b2] at [company_d3f4] was to reframe the roadmap conversation around the three constraints. Once we had alignment, the [amount_9e6f] budget question in [date_c8a2] resolved itself.

Red = entity detected by BERT-NER · Teal = deterministic placeholder · The reasoning survives. The people don't.

Who it's for

Built for people who use AI
as a thinking partner.

If you use AI to draft emails and book meetings, there's little to capture. If you use it to solve hard problems — this was built for you.

Consultant / Advisor
You think for a living. Your thinking should compound.
Client situations are similar enough that frameworks transfer — but without capture, you rebuild from scratch. IE turns every client engagement into a reusable reasoning asset.
Captures: reframe patterns · pricing objections · constraint-alignment approaches · client communication tactics
Product Leader / Strategist
Your decisions happen in AI threads. Your rationale disappears.
The decision was right. The reasoning that got you there — the tradeoffs weighed, the options ruled out — evaporates when the context window closes. IE persists it.
Captures: prioritization frameworks · tradeoff logic · stakeholder alignment moves · roadmap rationale
Researcher / Analyst
You synthesize constantly. Your synthesis isn't searchable.
You generate insight all day in AI conversations. None of it flows back into a searchable knowledge base — until now. IE captures the synthesis, not the source.
Captures: mental models · synthesis patterns · research frameworks · argument structures
Privacy architecture

Privacy is the product.
Not the policy.

These are commitments enforced in code. Not in a terms of service document you'll never read.

🛡

On-device. Always.

Source available on request · hussein@nextdecentrum.com

Raw conversations
Processed in-memory inside the offscreen document. Discarded immediately after anonymization. Never written to disk.
Never stored
Identifiers
Names, companies, locations, amounts, dates — all rewritten to placeholders. Any that survive verbatim are dropped, not stored.
Removed
Storage location
Your browser's IndexedDB. Inside your Chrome profile. We have no servers to receive anything because we have no servers.
On-device
Network calls
One-time model download from Hugging Face / jsDelivr on first use (software fetching itself). Zero user data in those requests. No further traffic.
Zero data sent
Account / login
None required. There is no account. There is no sign-in. There is nothing to authenticate.
None
Analytics / telemetry
None. Not anonymized aggregate stats. Not crash reports. Nothing. We have no way to observe your usage.
None
Delete
Per-entry, per-domain, or purge all — instantly. Uninstalling the extension deletes everything. No retention on our end because there is no "our end."
Instant
Research basis

Anchored in science.

Every design decision has a source we can point at. We built this because the research said we had to.

31×
A user tracked their re-research behavior for two weeks. They solved the same problems 31 times across AI sessions. Roughly six hours a week of intelligence that didn't compound.
User self-report · 2024 · Reddit thread on AI habits
−Critical thinking
Frequent AI use correlates negatively with critical thinking, mediated by cognitive offloading. The effect is strongest in the 17–25 cohort — but measurable across all ages.
Gerlich M. · Frontiers in Psychology · 2025
76%
Of AI-using professionals pay for at least one AI subscription. The person who needs an on-device intelligence layer is already paying for AI and already attuned to its value.
State of AI Report · 2025
85%
Chrome WebGPU coverage means 85% of Chrome users can run full on-device Llama-3.2-3B for classification and anonymization. The remaining 15% get a WASM fallback — nothing is blocked.
MDN caniuse data · July 2026
Pricing

Two tiers. Fully honest.

The free tier delivers the complete core loop — capture, anonymize, search, export. Pro is a power-user convenience layer, not a capability lock.

Free · Always
$0
No credit card. No trial. No expiry.
  • All four supported platforms
  • Semantic search across your full base
  • Up to 1,000 stored entries
  • Markdown + JSON + HTML export
  • Global · per-platform · per-conversation pause
  • PII-scan on every export
  • Domain browse (Pro)
  • Unlimited entries (Pro)
Everything you need to understand how Intelligence Engine changes how you work.
Pro · Beta
$12/mo
Pro features available free during beta.
  • Everything in Free
  • Unlimited entries
  • Domain browse — organize by topic
  • Chat with your intelligence (Phase 2)
  • Encrypted backup export (Phase 2)
  • Priority email support
  • Early access to new features
Beta participants get Pro features at no charge until the Chrome Web Store launch.
Private beta · Limited slots

Start capturing
how you think.

Join the private beta. We'll send install instructions as slots open. Pro features are free during beta — no card required.

Your email lands in the beta invite queue. Nothing else, ever.
No server. No sync. No cloud. Uninstall = delete everything. Pro features free during beta.