Barnaby Fox
"Thanks for the data! Your profile is complete."
Nitro Mode Unlocked
Earned by completing your profile
Chill
Nitro
⚠️

Light Sensitivity Warning

Nitro Mode contains rapid flashing effects, intense animations, and bright colors.

If you have photosensitive epilepsy, light sensitivity, or are prone to seizures, please do not activate Nitro Mode.

Nitro Mode significantly increases animation speed and visual intensity throughout the site.

CHANGELOG

Product Updates

Track every improvement, feature, and fix.
See how hk3k evolves to better serve your knowledge management needs.

⚠️
Chrome extension retired (March 2026). hk3k is now a web-only app at app.hk3k.ai — works in any browser, no install. Entries below from before March 2026 reference extension features that are no longer part of the product.
May 2026
Latest

Website Builder Backend

May 13
  • Contact forms — added end-to-end. Visitors submit, the site owner gets the email. Multi-recipient routing, HMAC-signed webhooks, CSV export, honeypot + per-IP rate limits all built in. Owners manage everything from chat: "send form submissions to bob@example.com", "show me my latest form submissions", etc.
  • AI chat with your site — added a chat bubble that visitors can open on any page and ask questions about the site itself. Answers are grounded in the site's actual content (per-site Pinecone namespace, SSE-streamed). Owner can rename the chatbot, set its persona, and customize the opener — all from chat.
  • KV store — added a tiny per-site named-value scratchpad. Toggle a "Spring Sale!" banner on/off by saying "turn on the spring sale banner", update "today's special" by saying "set today's special to lattes $5". No LLM re-write, the change is live within seconds.
  • Newsletter signup — added native list capture. Email-only forms are auto-detected as newsletter signups; double opt-in, one-click unsubscribe, CAN-SPAM postal address all handled. Owners see subscribers in chat: "how many subscribers do I have?"
  • Auto-wire helper that runs on every page save — new sites get forms, chat, KV, and newsletter wiring for free, and LLM edits never silently break existing wiring.

Marketing Site Refresh

May 1
  • Rewrote homepage to lead with workspace-first framing ("An entire workspace, one conversation away.")
  • Added "Already have a website? Bring it in." section with side-by-side import comparison demo
  • Demoted website builder from headline product to "one of hk3k's most-loved features"
  • Fixed broken og:image meta tag (was 404 sitewide); regenerated social-share image to match new framing
  • Reframed about, pricing, sites, and changelog pages for congruence with the new homepage
April 2026

Spaces, Agent Loop, Sites, Telegram

April
  • Spaces — unified HyperSearch + Projects into a single tree of folders, conversations, notes, tasks, files, and sites
  • Agent Loop — one prompt creates an entire nested workspace (folders, notes, tasks, conversations, even a published site), every action undoable
  • AI Website Builder — describe a site in plain English; Barnaby plans the pages, writes the copy, sources ethical imagery, and publishes at yourname.hk3k.ai
  • Site Import — bring an existing live site into hk3k by URL; edit it conversationally afterwards
  • Scheduled Tasks — due dates, recurrence, overdue states; reminders delivered in-app or via Telegram
April
  • Telegram Bot (@hk3k_bot) — chat with your second brain from any phone, get task reminders where you already are
  • Barnaby butler personality — optional on/off toggle, privacy-aware forget levels (0=normal, 1=no-LTM, 2=full-privacy)
  • Conversational shared-element propagation — edit a nav, header, or footer once and it updates across every page of a site
  • Custom domains for user-built sites with on-demand TLS
March 2026

Chrome Extension Retirement & Projects Beta

March
  • Retired the Chrome extension. hk3k is now a web-only app — sign in at app.hk3k.ai from any browser. Web app reaches more people, ships faster, and avoids Chrome Web Store review delays.
  • Launched Projects (later renamed Spaces) — phases 1–6 — folders, conversations, notes, tree navigation, drag-and-drop, per-Space instructions
  • Improved Notion sync — multi-user OAuth (each user connects their own workspace), Smart Capture indexing, incremental + full-sync modes
  • Background daemon thread architecture — search responses return in ~500ms even when LTM updates run in the background
February 2026

Infrastructure

February
  • Migrated from Render to a dedicated VPS (12 cores, 24GB RAM, 720GB SSD) — measurably faster end-to-end response times
  • Caddy reverse proxy with automatic Let's Encrypt TLS for hk3k.ai, app.hk3k.ai, api.hk3k.ai, and *.hk3k.ai user sites
  • Switched to async logging architecture — eliminated blocking file I/O from request paths
September 2025

LLM Observability & Search Optimization

September 16
  • Implemented Weave LLM tracing for comprehensive observability
  • Implemented intelligent internet search optimization for AI queries
September 15
  • Implemented auto-expanding input for HyperSearch with proper overflow handling
  • Implemented stacking placeholder spinners for pending search threads
  • Fixed Claude API attachment handling for images and documents
September 14
  • Implemented conversation cloning feature for HyperSearch
  • Fixed attachment handling in HyperSearch continue thread mode
  • Added cosine_similarity method to AIVectorizing class

GreaseRails Architecture & Search Enhancements

September 13
  • Implemented GreaseRails unified search architecture with major refactor
  • Implemented async background verification for search thread embeddings
  • Added import buttons to HyperSearch with dynamic visibility
  • Implemented hybrid keyword + semantic thread search with robust embedding creation
September 12
  • Implemented admin-based access controls across multiple pages
  • Limited conversation imports to 100 at a time
  • Added thread search bar to HyperSearch sidebar
  • Implemented PostHog analytics tracking across all frontend pages
  • Fixed feedback board sorting and added activity tracking
  • Added admin mode to feedback board with delete and status management
  • Added comprehensive feedback board system with user profiles
September 11
  • Added .vercelignore file to exclude unnecessary files from deployment
  • Added PostHog analytics integration for frontend tracking
  • Fixed Notion sync bug: use deep copy to prevent race condition

Authentication, Data Import & API Migration

September 10
  • Fixed Notion sync: OAuth authentication, page counting, and dual sync process
  • Fixed authentication security and added password reset feature
  • Fixed data reset functionality issues
  • Added user-friendly display names for Claude models in AI search receipts
  • Added Claude models to HyperSearch interface
  • Implemented complete LLM API migration architecture
September 8-9
  • Implemented complete Claude data import feature
  • Fixed ChatGPT import to display correct model names
  • Implemented complete ChatGPT data import feature
  • Removed numpy from requirements.txt to fix Vercel 250MB limit
  • Added neo4j package to requirements.txt

Graph Database & Document Processing

September 7
  • Added graph database viewer page with Chrome extension integration
  • Enhanced graph database with comprehensive temporal tracking
  • Fixed graph context retrieval for broad queries in AI search
  • Fixed missing graph context in continued AI search threads
  • Implemented comprehensive graph database integration with Neo4j Aura
September 6
  • Added Excel/XLSX support without pandas dependency - major milestone
  • Added intelligent routing for Word and text document attachments
  • Added intelligent CSV processing with hybrid context compression
  • Fixed .txt file attachment processing
  • Expanded HyperSearch to support multiple document attachments
September 5
  • Added required operation_type to LLM logging for better debugging
  • Used dedicated model for note updates based on eval results
  • Fixed AI search thread tag accumulation and note updates
  • Renamed AI Powered Search to HyperSearch
  • Moved clear button from bottom-left to top-right in search apparatus
  • Renamed SMI to Personalization in user-facing interfaces
  • Fixed AI search streaming text spacing with industry-standard CSS typography

Evaluation Framework & Optimization

September 2-4
  • Updated note generation to use v2_structured prompt (evaluation winner)
  • Fixed encoding issues in note generation test scripts
  • Updated Notion token validation to accept new 'ntn_' prefix format
  • Implemented chain-of-thought tag extraction with gemini-2.5-flash-lite
  • Added tag extraction evaluation framework with quality scoring
  • Implemented relevance check improvements with 2-shot prompt
  • Added tag hierarchy evaluation framework and scripts
  • Refactored search compression to use RecursiveSummaries
  • Optimized SMI updates with gemini-2.5-flash-lite model
  • Added comprehensive LLM evaluation framework for summarization and SMI extraction
September 1
  • Fixed asyncio import error and removed debug statements
  • Cleaned up repository - removed backup files and images to fix Vercel deployment
August 2025

Notion Integration & Smart Capture

August 21
  • Added Notion Setup interface for seamless user onboarding
  • Fixed Edit SMI button in floating interface
  • Enhanced AI search with SMI extraction capabilities
  • Implemented Smart Capture validation with database-only filtering
August 20
  • Redesigned bulk capture UI to match single capture modal
  • Added Bulk Notion Smart Capture feature
  • Enhanced Smart Capture with metadata extraction
  • Updated hotkey to Alt+Shift+R for better accessibility
August 18-19
  • Redesigned AI search interface with Claude-inspired two-row layout
  • Integrated Notion workspace data into AI search
  • Extended token expiration to 30 days for better user experience
  • Added Notion Data Explorer for admin workspace view

Authentication & Infrastructure

August 15-16
  • Implemented JWT-based authentication with Supabase Auth
  • Added Notion OAuth integration with secure state management
  • Fixed OAuth redirects for Chrome extension compatibility
  • Added dynamic API endpoint configuration for local/production
August 13-14
  • Added CORS support for Chrome extension
  • Implemented conditional file logging for Vercel deployment
  • Added configurable semantic chunking model
  • Optimized semantic linking for AI search

Memory & Tag Management

August 12
  • Implemented Pinecone-optimized tag streamlining system
  • Added intelligent tag merging capabilities
  • Updated extension popup with new branding
  • Added URL tracking and display for memories
August 11
  • Fixed tag hierarchy extraction for 551+ tags
  • Added hierarchy breadcrumbs to notes manager
  • Implemented memory receipt system with clickable references
  • Replaced emoji icons with clean SVG icons

AI Search Enhancements

August 8-10
  • Implemented multi-provider AI search with OpenAI and Google models
  • Added chat without internet search mode
  • Implemented AI search verbosity control
  • Added retry logic for Gemini responses
  • Fixed OpenAI image vision capabilities
  • Created distribution packages for non-technical users
  • Migrated to comprehensive token-based authentication
August 5-7
  • Implemented competitor-inspired dual-state search interface
  • Added multi-modal image support to AI-powered search
  • Implemented non-blocking UI for rapid operations
  • Added comprehensive hotkey system
  • Enhanced citation system with [SOURCE X] format
Early August 2025

Notes Manager & UI Improvements

August 3-6
  • Implemented soft delete archive system for Notes Manager
  • Added semantic search functionality to Notes Manager
  • Implemented automatic tag hierarchy organization
  • Fixed Pinecone eventual consistency issues
  • Added comprehensive notes management with AI generation
August 2-4
  • Added modern dark/light mode toggle across all interfaces
  • Implemented clickable search thread titles in notes
  • Added comprehensive search thread management
  • Simplified Chrome extension popup to navigation-only
  • Added pagination to AI-powered search threads
July 2025

Foundation & Core Features

Late July
  • Added AI-Powered Topical Report feature
  • Implemented todo management functionality
  • Added comprehensive tagging system
  • Enhanced memory manager with date sorting and SMI filtering
Mid July
  • Added complete memory manager interface
  • Implemented system message inserts UI
  • Added keyboard shortcuts to Chrome extension
  • Fixed vector ID algorithm with timestamp-based unique IDs
July 17 - Initial Launch
  • Initial commit: Hyper Knowledge project launched
  • Combined Total Recall memory system with AI-powered search
  • Core Chrome extension functionality
  • Basic API infrastructure

Stay Updated

Get notified about new features and updates as soon as they're released.

Get Started Free View Roadmap