One Year of $CHAT! 🎉December 19th marked VectorChat's first anniversary! Our evolution has been remarkable: from pioneering web3 conversational AI to directly competing with Character.​ai and Chai for market leadership.Chunking, now a core part of our project, first emerged as a challenge, where we wanted to deliver the same immersive experience to a much larger and predominantly free audience.What started as an internal solution grew into a Bittensor $TAO subnet that has now surpassed industry leaders @AI21Labs and @UnstructuredIO in most contexts—delivering the world's best chunking technology!Some major Toffee breakthroughs slashed costs while maintaining excellent quality:🔸 Images: $0.00025 each (~200x lower than commercial APIs)🔸 Voice: $0.025 per hour (~20x lower than commercial APIs)While others focused on improving user retention:🔸 Cost-effective use of RAG for unparalleled character memory and immersion🔸 Dynamic Emotions and Backgrounds—the character physically evolves with the conversation🔸 Candies: Immersive worlds to seamlessly swap onto your charactersWith Toffee AI and Chunking.​com just around the corner, next year looks incredibly bright!

One Day until VectorChat x @partyhatDAO AMA!Remember to tune in tomorrow, December 13th at 6pm EST, to hear from us and our wonderful hosts!Submit questions either here or to the quoted tweet.

VectorChat x @partyhatDAO AMAJoin us December 13th at 6pm EST for our first ever spaces!Hosted by @naniXBT, @CoinGurruu, and @toptickcrypto.Spaces link: https://t.co/iWYLeU3uasSubmit your questions here! 👇

What does our team look like now?As the VectorChat project has grown and evolved, so has our team! Our philosophy: small team, exceptional talent. 🚀 Toffee (permanent, full-time)- UI/UX Designer- Mobile App Developer- Frontend Web App Developer- Full-Stack Developer- AI Infrastructure Developer (PhD)Experts- GA4 Analytics Expert + Data Scientist- Voice TTS model expert- LLM prompting expert- LLM fine-tuning expert- SillyTavern & Lorebook expert- Image fine-tuning expertChunking (permanent, full-time)- Full-Stack Developer (Subnet)- Full-Stack Developer (Platform)- NLP/RAG Researcher (PhD)- NLP/RAG Researcher (Georgia Tech)

🌐 What is VectorChat?As we approach our one-year mark, we're excited to reintroduce VectorChat to our growing community!The VectorChat project consists of multiple divisions, all ultimately towards the success of the $CHAT token.Divisions of VCAI:🍬 Toffee: An uncensored, powerful conversational AI platform set to drive early demand for $CHAT. Toffee is also a large client of https://t.co/GOlUoQCQXp.🧩 https://t.co/GOlUoQCQXp: An enterprise platform that will offer unmatched chunking powered by our $TAO subnet, with plans to expand into reranking, embedding, and retrieval.🗣️ $CHAT: Currently a speculative token, $CHAT will evolve into a decentralized protocol, initially fueled by demand from Toffee before scaling to attract other permissionless clients.Look forward to deeper dives into each component of the project!

Lucid by VectorChatWe're excited to announce the launch of Lucid, a robust and modular solution to eliminate relay mining in Bittensor!By leveraging a private IPFS cluster, inscriptions, and cryptographic verification, Lucid provides the tools to allow subnets to tackle relay mining with ease.Learn more about the Lucid architecture:https://t.co/7xVsEirvnRSee how we’ve applied it to the @chunking_subnet:https://t.co/tGOo0K0CYv

Are You A Cracked Dev?We're expanding our team at VectorChat with three key leadership roles for our cutting-edge AI projects: Toffee and https://t.co/GOlUoQDoMXPassionate about advancing conversational AI or data processing? Apply now!See jobs: https://t.co/lRCLUw8PMH or @XHiring

🗺️ Subnet Development RoadmapPhase 2 of the VectorChat project—Decentralization—is now in full swing as our subnet launches on Bittensor!🔗 Attached is the roadmap for the subnet, incentivizing the development of the premier chunking solution!Immediate ➡️ Demand ➡️ Multimodality

VectorChat $CHAT: 7/3-7/19 Dev Log Web3: - Registered subnet on $TAO mainnet - Benchmarked our service against various chunking services (e.g., @UnstructuredIO, @AI21Labs, @LangChainAI) - Created SDKs for python and Node - Outlined example project code - Planned content for user, validator, and $CHAT dashboards - Designing landing page Web2: 7/3-7/11: - Benchmarked performance and cost for various models on… + Different hardware configurations + Different parameter size + Varying degree of Quantization - Tested OpenVoice v2, xtts v2, StyleTTS - Testing vLLM, optimizing KV cache for cost - Migrated backend to python - Switched hybrid vector to use bert sparse embedding - Finished integrating candy creation, candy discover, and candy packs pages - Added custom usernames and profile pictures to profile pages - Various UI adjustments and fixes to chat page - Implemented backend for Voice 7/12-7/19: - Upgraded chat endpoint - Finished candy creation page - Finished search by tag/character/user/candy - Added AI generated images and descriptions to character creation flow - Updated RAG pipeline for lower latency filtering by candy metadata - Create tag/category discover pages - Added tag/category to character creation - Updated profile page with edit profile, like profile, private/public

🛠️ VectorChat Primary Site OverhaulWe’ve updated our site to be the central hub for all things VectorChat!Dive into detailed explanations of every key project component:✨ Project Structure✨ $CHAT Token and Utility✨ Chunking Subnet and https://t.co/GOlUoQCQXp✨ The Toffee Platform✨ Updates & Career OpportunitiesAnd access all our resources in one place!🔗 Explore the new VectorChat: https://t.co/tJdrnfswd1

VectorChat $CHAT: 6/27-7/3 Dev Log Web3: - Benchmarked our service against various chunking services (e.g., @UnstructuredIO, @AI21Labs, etc.) - Created SDKs for python and Node - Outlined example project code - Planned content for user, validator, and $CHAT dashboards - Designing landing page Web2: - Benchmarked performance and cost for various models on… + Different hardware configurations + Different parameter size + Varying degree of Quantization - Tested OpenVoice v2, xtts v2, StyleTTS - Testing vLLM, optimizing KV cache for cost - Migrated backend to python - Switched hybrid vector to use bert sparse embedding - Finished integrating candy creation, candy discover, and candy packs pages - Added custom usernames and profile pictures to profile pages - Various UI adjustments and fixes to chat page - Implemented python backend for voice

VectorChat $CHAT: 6/20-6/27 Dev Log Web3: - Updated README - Began work on designing https://t.co/GOlUoQCQXp landing page - Began work on Chunking Dashboard for validators - Began work on https://t.co/GOlUoQCQXp Dashboard for customers - Designed preliminary system for $CHAT access to subnet - Developed SDK cores - Created API key system for customers - Revised miner selection for validator - Standardized formatting of repo Web2: - Worked extensively on Candies + Added endpoints for creation, deletion, querying, and forking + Implemented candy creation page + Implemented candy discover page - Created forgot password, sign up, set profile picture and name - Migrated character creation flow to new design - Integrated eleven labs voice demo - Designed and implemented new chunking algorithm - Began benchmarking RAG performance

Structure of the VectorChat ProjectA brief look at a graphic from our upcoming landing page, going over the core components and their interoperability. Core Components:✨ Toffee: A cutting-edge conversational AI platform setting new standards in user interaction.✨ https://t.co/GOlUoQDoMX: Delivers innovative chunking solutions for developers and enterprises powered by our subnet.✨ Chunking Subnet: Incentivizes the creation of the most sophisticated chunking solutions, enabling superior retrieval augmented-generation across text, audio, video, and raw data.✨ $CHAT Decentralized Protocol: The ultimate goal, incentivizing and accessing decentralized intelligence surpassing that of centralized systems.

VectorChat $CHAT: 6/12-6/20 Dev Log Web3: - Developed core of task API (validator bandwidth monetization network) - Secured task API with digital signatures - Added logging system - Added mechanism for validators to receive organic requests - Added task API options to validator config - Added generate synthetic task function to validator code - Added task queue to task API - Designed and implemented mechanism for assigning ids to tasks Web2: - Completed migration to NextAuth - Added Edge dependency for NextAuth - Implemented search + ranking endpoints - Fixed implementations to be consistent with design - Setup custom categories + tags - Completed character + candy (knowledge pack) creation pages - Migrated creation page functionality - Updated Chat UI

VectorChat $CHAT: 6/04-6/12 Dev Log Web3: - Tested on testnet - Created VectorChat channel on Bittensor discord - Designed and implemented relative ranking system - Revised incentive curve - Revised query pattern - Fixed issue with neuron communication Web2: - Began implementing designs for Discover page (Look forward to previews/demos!) - Finished implementing hybrid embeddings using OpenAI - Implemented chat page, recent chats, and regeneration in new UI - Implemented hybrid search using keyword matching in new UI - Implemented character search by name in new UI, as well as ranking algorithm for additional recommendations - Added support for custom categories, building the foundation for user tagging - Designed and implemented search page - Designed and implemented profile page - Began NextAuth migration for new profile page

VectorChat $CHAT: 5/25-6/03 Dev Log Web3: - Tested on localnet - Launched on testnet - Revised reward mechanism - Revised config options - Revised protocol - Hardened default miner against spam - Researched product-market fit for chunking / RAG - Designed landing page Web2: - Finished implementing discover page - Implemented new sidebar - Finished recommendation algorithm + Suggestion based on user activity + Aggregation based on subscriptions, messages, etc. - Finished implementing support for multiple categories - Updated schema to support multiple categories - Implemented combobox component - Updated character creation flow to support multiple categories - Designed NextAuth migration

🌐 The Chunking Subnet TL;DR of the TL;DR:✨VectorChat is developing a subnet on Bittensor✨Subnet handles Retrieval-Augmented Generation (RAG)✨Subnet is starting with chunking✨Our RAG already outperforms OpenAI✨Effective chunking will become more critical as data grows exponentiallyRead about our subnet and what it means for VectorChat here: https://t.co/BVy4y6Zec5

VectorChat $CHAT: 5/16-5/24 Dev Log Web3: - Identified alignment between needs for the new platform and the capabilities of Bittensor $TAO + A detailed description will be provided in the very near future! - Re: Subnet: + Developed and are finalizing design for subnet + Designed and coded incentive mechanism + Designed and coded reward mechanism + Designed and coded algorithm for the generation of dummy data (for when no requests) + Developed protocol for neuron communication between validators and miners + Developed Default Miner + Developed Default Validator - Working on documentation and new static site Web2: - Revamped discover page + Implemented new themed UI + Implemented new character + knowledge pack cards, tags, recent chats, and more - Added recommendation algorithm to provide characters based on user preferences - Implemented UI for new chat page + Show instances of chats + Regenerate, evolve button + Settings for custom backgrounds - Allow users to create custom categories - Add ability to chat with multiple instances of a character - Designed all UI components for new platform - Began working on implementing new search - Added Knowledge Pack search and deletion to backend

VectorChat $CHAT: 5/09-5/15 Dev Log - Updated model to to GPT-4o - Updated schema to support further distinction for utility characters - Removed double generation of canceled streamed messages - Fixed carousel responsiveness - Fixed responsive grids for companions - Finished message regeneration - Improved markdown/latex parsing responsiveness - Improved recommendation algorithm (not live) - Improved backend metrics for evaluation - Implemented scaffolds for new design - Modified context parsing for image/video upload - Expanded layout size on evolve page - Refactored chunking and indexing algorithm - Switched to cosine similarity for vector embeddings, allowing implementation of keyword vector search

GPT-4o is now integrated into the VectorChat Beta! https://t.co/sUMafYN0lJ

VectorChat $CHAT: 4/25-5/08 Dev Log For front-end / new platform: - Revised UI/UX design for Chat component - Designed Search component - Designed Recommendation component - Designed Profile component - Designed Settings modal - Designed Chat Customization modal - Began work on designing Character Creation Flow For backend / legacy platform: - Added image support to ‘Evolve’ page using OpenAI - Added video support to ‘Evolve’ page with youtube api + w/ summary - Added hybrid keyword search using BM25 sparse embeddings - Minor styling + code restructuring of ‘Evolve’ page - Fixed spacing, margin, and chat overflow issues - Reverted back main card styling and started design implementation for auth migration - LlaMa 3 testing on groq for lightning fast inference - LlaMa 3 deployment migration to Anyscale - Fixed LaTeX single dollar rendering issue - Began work on recommendation + custom search algorithm

VectorChat $CHAT: 4/18-4/24 Dev Log - Added support for uploading images in evolve page - Added support for regenerating a bots response - Added support for custom background images (can show example) - Added support for @GroqInc / updated to Llama 3 - Added support for direct video upload in ‘Evolve’ page (now working on upload via url) - Added settings page and modal (deciding) - Blueprinted Knowledge Pack version history algorithm - Developed recommendation algorithm based on past user history - Designed chat page for UI rebrand - Updated UI to include chat bubbles - Updated trending algorithm with a threshold on interactions to avoid showing unpopular companions - Updated dynamically sized card size, margins, and centering - Upgraded pinecone vector databases to optimize for performance and add support for keyword search with sparse vectors - Refactored code to follow best practices

VectorChat $CHAT: 4/10-4/17 Dev Log - Dynamically size character cards to support a variety of devices without weird gaps in display - Dynamically switch pdf rendering to client-side if it results in an optimization - Updated statistics algorithm for trending companions - Updated indexing algorithm with flexible support for data streams (tweets, messages, etc.) - Updated image UIs to support transparent and large dimension images - Implemented new graph-based memory retrieval algorithm giving characters more nuanced memory - Designed UIs for image and video upload - Switched to OCR PDF parser (and begin developing image support) for PDF parsing - Improved knowledge pack search algorithm in character creation flow - Added loading skeletons for various components - Added customization to character creation flow to support - Added security rules to production environments to prevent random crashes or buggy updates - Added dynamic loading to components to prevent SSR - Planned migration from clerk to next-auth

VectorChat $CHAT: 4/3-4/10 Dev Log - Recruited four incredible talents - Setup company accounts and tools for newly onboarded devs - Brought new devs up to speed to hit the ground running - Added latex rendering for bot responses - Added support for multi line inputs on mobile - Added loading components for chat and characters - Modified pdf parsing to OCR library for better results - Fixed user profile display on mobile devices - Setup framework and UI design for TBA Knowledge Pack feature - Improved text cleanup for PDF parsing - Improved load times for user accounts in chat window - Improved internal benchmarking for vector database retrieval and indexing - Implemented backend infrastructure for hybrid vector embeddings

✨ $CHAT: Build-In-Public ✨As outlined in our Medium article, now referred to as CHAT001, the VectorChat brand is set to bridge the gap between our team and the Web3 community.A key aspect of this commitment is our "build-in-public" approach, where we’ll consistently share updates on our platform’s progress—be it technical, milestone-based, or creative—as they unfold.Today, we offer a sneak peek at our character creation flows. Below is a tentative design for the “Upload Knowledge” page within the Utility bot creation process, shown in both light and dark modes.

🎖️ $CHAT Community ContestDemonstrate your technical ability and creative prowess!Craft an outstanding utility character on VectorChat, whether it’s a wellness mentor, coding guru, emotional support guide, or any other concept you dream up!The top three submissions will win $1250, $750, and $500 of $CHAT. The next five will earn $100.Like, RT, and respond to this post with the link to your character to be eligible for the contest. You may enter multiple characters.🏆 Winners to be decided Apr 30th

🔭 The Future of VectorChat TL;DR of the TL;DR: ✨ We're all in, and we're going for the king ✨ Restructuring and Rebrand ✨ $CHAT: A Decentralized Inference Protocol ✨ Marketing Overview Read about VectorChat's future in our Medium article: https://t.co/Aow8sfxPyV

✨ Utility Update Preview ✨As VectorChat gears up to dominate the conversational AI landscape, we are undergoing a complete site revamp, both refactoring existing code for enhanced performance and scalability, and also building in real, yet streamlined utility, making our platform intuitive and powerful.One such change is a total rework on how users create models. Gone are the days of facing a daunting array of broad, open-ended questions! Now, the site will guide you through the model creation process, simplifying it even if your idea is still nascent.Attached below is a sneak peek at the first step in the new creation process. The flow will adapt based on the model type you're aiming to develop, with 🪄 AI assistance available at every juncture! Stay tuned for more updates as we continue to innovate in the open!

VectorChat $CHAT: 3/13-3/20 Dev Log - Added multi-line inputs in chat window - Added rendering of markdown and code - Added infrastructure for setting usernames and eventually creating entire profile customization page - Added support for converting uploaded images for characters and profile pictures to both png and jpeg - Better formatting of character outputs like separation into logical paragraphs - Implemented support of transparent images for sharing - Fixed search functionality, and improved search to use searching algorithm matching entire substrings rather than only prefixes - Created on-boarding documents & updated internal documentation for new developers

VectorChat $CHAT: Utility Update!As we continue on our mission to build the best conversational AI platform out there, we will start rolling out utility updates, or minor updates designed to make the platform either easier to use, more stable, or to increase the potential of our users. The first of many in this series is Programming! Now, you can code alongside the likes of Mark Zuckerberg or Bill Gates, or you can design and build utility-oriented characters to help you program. Whether you're a seasoned coder or just starting out, this update opens up a world of opportunities to expand your coding prowess.

✨Surprise Sneak Peek ✨A couple months ago we announced and delivered multi-modality, even though it wasn’t on our roadmap. Today, we have another surprise announcement: programming!Designed to enhance utility, VectorChat characters will soon be able to respond to you with code, in an easy-to-use and interactable UI.Ever wanted to code with Mark Zuckerberg?

$CHAT with characters using images on VectorChat! https://t.co/X7TaByJuhf

VectorChat $CHAT Beta Update! The long awaited Public & Private Sharing Update is finally here! Share your creations with others, either privately with your friends, or publish them publicly to share with our community! This update opens the door for our new and enhanced roadmap, including: ✨ Revenue sharing for token holders! ✨ Monetization for AI creators ✨ New and innovative mechanisms for viral content generation 👀

VectorChat $CHAT: 2/21-2/29 Dev Log - Further optimizations to evolve algorithm, halving compute resources, allowing us to double the number of storage per character to 10 million tokens - Add ability to favorite companions from the dashboard - Add metadata to sharing page, allowing more aesthetic sharing - Add "my profile" page - Double limits for pdf sizes through various text cleaning and backend optimizations - Finalize trending and popular algorithm - Add image and chat moderation - Modify sidebar ui to include contact info - Modify category selection ui - Mobile friendliness ui changes for carousel in discover page and profile pages - Implement various checks related to file upload (duplicate files, invalid file types, etc.) - Upgrade embedding model from ada-002 to text-embedding-3, slightly improving evolve algorithm - Change hover ui on cards to display sharing visibility - Small redesigns to various core ui components (pop ups, dialog, buttons) - Modify image upload backend to allow exporting character images to various image formats - Add popup dialogs to various systems that would benefit from warnings like publishing characters to discover page