This spreadsheet is a simple database of selected AI hardware, software and models to inform community decisions around planning and deployment of AI inference platforms for the IC. The collected information has so far largely come from me but I am keen to get everyone who is interested contributing to this as a shared resource. Feel free to request edit access then read, add and comment on anything.
Hi everyone, thank you for today’s call (2025.06.19). Special thanks to @icarus for leading the call! This is the generated summary (short version, please find the long version here ): The DeAI Working Group is developing a structured model selection framework to identify LLMs best suited for ICP AI use cases and deployment. By categorizing model families and variants by size, architecture, licensing, runtime compatibility, and deployment feasibility, the group aims to match model capabilities—ranging from simple tool use (tiny models) to complex multi-agent planning (large models). Special emphasis is placed on open-source models like OLMoE and Gemma, support for lightweight runtimes like GGUF and vLLM, and the creation of a shared benchmark and spreadsheet tool to guide the community in choosing the right model for decentralized deployment on ICP.
Links shared during the call:
- AI4AI Infrastructure Options Analysis: https://docs.google.com/spreadsheets/d/1a-NzZeh6D9smxovW7ByYOx7Ls1tRqBJyaS9gXVzRy-Y/edit?usp=sharing
- Recommended AI blog: https://www.interconnects.ai/
- Research on LLM Size vs. Agentic AI Capabilities: LLM Size vs - Google Docs
- https://allenai.org/
- Introducing the LLM Canister: Deploy AI agents with a few lines of code - #54 by ddave
- https://ai.meta.com/blog/llama-4-multimodal-intelligence/
Hi everyone, this week we’ll have an exciting demo by Sam on the plugin he built that enables ICP to be integrated into the Eliza AI framework
I’m looking forward to the session!
Hi everyone, thank you for today’s call (2025.06.26). Special thanks to Sam for sharing his work! This is the generated summary (short version, please find the long version here ): Samuel demoed a powerful new plugin for the Eliza AI agent framework that enables interaction with the Internet Computer (ICP), including wallet management, token swaps, and full NNS neuron control. The modular plugin architecture separates actions, providers, and evaluators, and integrates smoothly via Eliza’s character files. With built-in safety features like two-step confirmations, the plugin sets the stage for future capabilities like SNS proposal analytics and autonomous investment insights. Community feedback highlighted it as a major step toward seamless agentic UX on ICP.
Links shared during the call:
- GitHub - sam-thetutor/plugin-icp: An Internet Computer Protocol (ICP) plugin for Eliza OS, enabling meme token creation, ICP canister interaction, and wallet management.
- https://www.elizaos.ai/faq
- https://www.elizaos.ai/
- https://medium.com/ai-dev-tips/create-ai-agents-with-ai16z-in-15-minutes-639751e6ea69
- https://arxiv.org/pdf/2501.06781
- Add Model Context Protocol (MCP) Support · Issue #844 · elizaOS/eliza · GitHub
- https://www.youtube.com/watch?v=UAQMM95TDOY
Hi everyone, this week we’ll have an exciting demo by Sam from BIPQuantum (https://x.com/bipquantum) who are merging AI and ICP to tackle IP law! I’m looking forward to the session ![]()
Hi everyone, thank you for today’s call (2025.07.03). Special thanks to @baolongt for demoing his latest work on MCP and @Jupes_M for sharing his work on AI residents in the metaverse! This is the generated summary (short version, please find the long version here ): In today’s DeAI Working Group call for the Internet Computer, Long gave an extensive live demo of the MCP server running on IC, showing how AI agents can perform wallet operations (like balance checks, ICP transfers, and transaction histories) via HTTP gateways, while also discussing current authentication challenges. He highlighted progress on multi-chain wallet support with threshold signing for Bitcoin and Ethereum. Meanwhile, Jupes shared his work on running LLMs on a small Kubernetes Raspberry Pi cluster to power decentralized virtual AI “residents” in a simulation-style metaverse. The group explored ideas around integrating these with personal MCP servers, discussed potential for AI to propose rules in such environments, and Long asked for further feedback on the MCP SDK by the community.
Links shared during the call:
- GitHub - ByteSmithLabs/mcp-examples: A repo of some Model Context Protocol servers built on Internet Computer Canister.
- share feedback on MCP SDK with Long on X: https://x.com/long2408
- GitHub - ilbertt/ic-react-native-jwt-auth: React Native (Expo) + JWT Authentication + Rust ICP canister
- GitHub - nobodywho-ooo/nobodywho: NobodyWho is a plugin that lets you interact with local LLMs, we currently support Godot and Unity, with even more plugins on their way.
- Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task — MIT Media Lab
- [2506.08872] Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task
Hi everyone, this Thursday we’ll have the demo by Sam from BIPQuantum (https://x.com/bipquantum) who are merging AI and ICP to tackle IP law! I’m looking forward to an exciting session and seeing everyone ![]()
Hi everyone, thank you for today’s call (2025.07.10). Special thanks to @SamDrissi and the BIPQuantum team for sharing their work! This is the generated summary (short version, please find the long version here ): In today’s call, we had a demo by the BIPQuantum team. BIPQuantum is an ICP-based platform that streamlines intellectual property (IP) registration and protection by minting IP as blockchain-based NFTs and issuing certificates, with future plans for AI-driven automation, royalties, and integration into US corporate certification workflows. Built on a multi-canister architecture, it currently supports small on-chain LLMs for IP-related queries, while exploring off-chain and hybrid approaches to overcome ICP’s memory and instruction constraints. With a user base of scientists, AI creators, and entrepreneurs, the project also aims to fractionalize IP for funding innovation. Early adoption of blockchain IP recognition in jurisdictions like France and China supports its vision, as the team actively develops voice and AI integrations to enhance decentralized, quantum-resistant IP management.
Links shared during the call:
- bipquantum | Twitter | Linktree
- BIPQUANTUM | Dapp
- GitHub - Recursive-Emergence/RE: the original ideas from Isaac
- Notion
- GitHub - onicai/llama_cpp_canister: llama.cpp for the Internet Computer
- Introducing the LLM Canister: Deploy AI agents with a few lines of code - #62 by dawnnguyen
- DeAIWorkingGroupInternetComputer/Projects at main · DeAIWorkingGroupInternetComputer/DeAIWorkingGroupInternetComputer · GitHub
- GitHub - patnorris/DecentralizedAIonIC: DeVinci is the browser-based AI chatbot app served from the Internet Computer. You can chat with the AI model loaded into your browser so your chats remain fully on your device. If you choose to log in, you can also store your chats on the Internet Computer and reload them later.
- https://devinci.onicai.com/
- https://icgpt.onicai.com/
Thank you @patnorris and the DeAI Working Group for the opportunity to present BIPQuantum during yesterday’s call ![]()
To the ICP dev community:
BIPQuantum is now live and ready for testing! We are building a decentralized, quantum-resistant platform for certifying, indexing, and eventually monetizing IP using multi-canister smart contracts + embedded LLMs.
What we’ve built so far:
- Fully functional IP registration & minting as NFTs
- Dynamic certificate generation
- On-chain LLM queries for IP-related Q&A
- Open roadmap for voice integration, IP fractionalization & marketplace logic
- Deploying next: open-source LLM canister (off-chain fallback ready)
We’re actively onboarding creators, AI engineers, and open-source contributors.
Current Airdrop: 1000 $BQC tokens to every new IP creator
DAO will reserve 1% of supply for early adopters
Try it here: https://dapp.bipquantum.com/login
Join our dev & user community: bIP Quantum Community
Or DM me directly if you’d like to test the platform or get involved.
Let’s push the edge of AI + IP on-chain together.
Sam Drissi
Founder, BIPQuantum
Hi everyone, @icarus will lead this week’s call on AI4AI where we focus on hardware and infrastructure for AI on ICP.
We’ll continue our initiative to analyze options for GPU hardware, AI frameworks/runtime and LLMs to run as collected in this spreadsheet: https://docs.google.com/spreadsheets/d/1a-NzZeh6D9smxovW7ByYOx7Ls1tRqBJyaS9gXVzRy-Y/edit?gid=712966265#gid=712966265
If you have any insights, options or findings you’d like to add, please go ahead ![]()
See you all on Thursday
Hi everyone, thank you for today’s call (2025.07.17). Specials thanks to @icarus for leading it! This is the generated summary (short version, please find the long version here ): Today’s DeAI Working Group call explored the rapidly evolving landscape of decentralized AI infrastructure, highlighting breakthroughs in open-source LLMs (like Kimi K2 and Phi-3), 4-bit quantization methods (HIGGS), and efficient inference hardware such as NVIDIA’s Blackwell GPUs and analog accelerators from Axelera and EnCharge. vLLM emerged as the dominant model-serving framework, with growing support for multi-modal and CPU-only inference. The group discussed how community members could self-host inference runners and contribute to a decentralized registry of model runtimes, as the DeAI ecosystem moves toward fast, modular, and self-sovereign AI execution.
Links shared during the call:
- https://vram.ai/
- https://www.youtube.com/watch?v=IST4ME7nfZk&feature=youtu.be
- https://github.com/Sid31/team-ai/tree/main
- Kimi K2: Open Agentic Intelligence
- HIGGS
- LLMs No Longer Require Powerful Servers: Researchers from MIT, KAUST, ISTA, and Yandex Introduce a New AI Approach to Rapidly Compress Large Language Models without a Significant Loss of Quality - MarkTechPost
- https://axelera.ai/
- https://www.enchargeai.com/
- https://en100.enchargeai.com/
- https://docs.google.com/spreadsheets/d/1a-NzZeh6D9smxovW7ByYOx7Ls1tRqBJyaS9gXVzRy-Y/edit?gid=712966265#gid=712966265
Hi everyone, in today’s session we’ll have an exciting demo by the DataPond.ai team! See you all in the ICP Discord’s voice channel at the usual 3pm ![]()
Hi everyone, thank you for today’s call (2025.07.24). Special thanks to Emiliyan from DataPond.ai for demoing their work! This is the generated summary (short version, please find the long version here ): During today’s DeAI Working Group call, Emiliyan Enev presented DataPond.ai, a decentralized AI-data marketplace built on ICP, emphasizing transparency, auditability, and monetization through blockchain-tracked data usage, quality points, and future integrations with NFTs and cross-chain interactions. The group discussed potential standards (like MCP), challenges around data protection, scalability, leveraging Bitcoin for payments, and strategic roles for ICP in multi-chain interoperability. The discussions also explored using WebAssembly (WASM) to optimize AI inference by eliminating inefficiencies like Python and enabling lower-level, economical runtimes suited for edge computing, particularly within the ICP ecosystem.
Links shared during the call:
- DataPond Canisters:
- Storage: GitHub - ReCheck-io/datapond-storage-canister
- Tracing: GitHub - ReCheck-io/datapond-tracing-canister: DataPond Tracing canister is a Internet Computer smart contract for seamless data transparency and accountability.
- Points: GitHub - ReCheck-io/datapond-points-canister
- Repo with the ICP integration: GitHub - ReCheck-io/icp-datapond-integration: This repository showcases illustrative diagrams and sample codes demonstrating the implementation
- Good talk about using wasm for LLMs: https://www.youtube.com/watch?v=upNNI_b4tNY
Hi everyone, @patnorris
I would love to present ICP Agent Kit to you in next meeting if possible. ICP Agent Kit is a comprehensive TypeScript SDK that brings natural language processing to Internet Computer blockchain operations.
You can see details here: ICP Agent Kit: Natural Language Interface for Internet Computer (LangChain Based)
GitHub: GitHub - justmert/icp-agent-kit
Docs: Introduction - ICP Agent Kit
Excitedly waiting to deep dive into details!
Hi @justmert , thank you for reaching out. A demo sounds great! We have another demo scheduled for this Thursday already, so would next week (Aug 7) also work for you?
Hi @patnorris, that’s okay for me.
Excitedly waiting to meet!
awesome, cheers, looking forward to it!
We usually recommend keeping the presentation/demo at 20min max to have enough room for questions and discussion afterwards. The format is up to you ![]()
Hi everyone, thank you for today’s call (2025.07.31). Specials thanks to QryptiQ for the demo! This is the generated summary (short version, please find the long version here ): During today’s DeAI Working group call, the IQube Protocol was demoed. IQube is a foundational system for building reliable, agentic AI by tokenizing data into NFT-like “iQubes” that bundle public metadata with private, encrypted payloads. These iQubes enable trust, verifiability, and risk-based valuation across decentralized AI systems. The protocol supports cross-chain interoperability, token-gated access control, and agent integration via the Agent Protocol, while the COYN layer introduces proof-of-risk-backed tokenization. A live demo, Aigent Nakamoto, showcases how users can permission AI agents to access personal data for hyper-contextual responses. With planned integration into ICP and Bitcoin for secure settlement and a focus on both B2B and consumer applications, IQube aims to lay the groundwork for a modular, censorship-resistant, and composable AI-data economy.
Links shared during the call:
- iQube Whitepaper PDF: DocSend
- Aigent Nakamoto: https://nakamoto.aigentz.me/
- Nakamoto Activation Guidde: Activate Your Aigent Nakamoto Account
- Aigent Z (Requires Polygon Sepoilla tokens to mint): https://alpha.aigentz.me/
- A language model built for the public good | ETH Zurich
Hi everyone, we’ll have an exciting demo by @justmert on ICP Agent Kit this Thursday! If you like, take a look at the links on this new SDK in the original message I’m replying to
Looking forward to the demo and seeing everyone then
Hi @patnorris,
Where can I see the meeting link and time for event?