Open beta is staring next week (we have closed beta now for anyone who asks to try it out). The plugin will be dropped into our OC with directions on usage. Based on the results and feedback we will post to Chome store shortly after.
A little bit of rifting on succinct verification for AI.
Kinic team is pretty close to finishing SOTA for zero-knowledge machine learning (2~ months or so). This means succinct proofs for many types of models run off-chain - with decent prover time. Not sure how well this will work for large LLM, but we plan to release benchmarks in any case. sub minute off-chain inference is a near term target. Other models like gradient boosting and prediction (DeFi) should be much faster.
The IC right now uses TEE in a special subnet for LLM inference. We plan on introducing zkML as another method that also supports private weights / models. We will market this to grab the attention of people in the larger AI space.
VetKey integration progress is good.
We would love to chat about SOTA on zkML and what it means in the context of the IC.
Verifiers on chain that can trigger other chains via Chain Fusion.
Hi everyone, thank you for yesterday’s call (2025.05.01). Special thanks to @swift for sharing his work on ANIMA! This is the generated summary (short version, please find the long version here ): In this session of the DeAI Working Group, Michael presented ANIMA, a project focused on building emotionally aware, persistent AI personalities (ANIMAs) on the Internet Computer. The group explored its architecture—combining LLMs, memory, emotional modeling, and on-chain identity—as well as its broader implications for AI cognition, personality simulation, and autonomy. Discussions ranged from technical challenges like HTTP outcalls and multimodal tool ecosystems to deeper philosophical questions about anthropomorphizing AI and simulating time, with participants recognizing ANIMA’s potential as both a user agent platform and research testbed.
Hi everyone, thank you for today’s call (2025.05.08). Special thanks to @apotheosis for the great presentation on zkML and the update on Kinic! This is the generated summary (short version, please find the long version here ): @apotheosis gave an update on Kinic’s browser plugin which is now in open beta and then deep dived into Zero-Knowledge Machine Learning (ZKML). His presentation explored ZKML on the Internet Computer (ICP), highlighting how zk-proofs enable verifiable AI computation without exposing sensitive data. He compared ZK with Trusted Execution Environments (TEEs), outlined technical advancements like folding schemes, Jolt-based ZKVMs, and WebGPU acceleration, and showcased real-world use cases such as private LLM inference and AI agent coordination. The ICP’s architecture allows for native on-chain verification of modern proof systems, positioning it as a powerful platform for decentralized, privacy-preserving AI.
Hi everyone, join us this Thursday to discuss and learn more about two of the hottest topics in AI these days: the Model Context Protocol (MCP) and the Agent2Agent Protocol. @baolongt will talk about his latest work on building MCP on ICP See you all then (the usual 6pm CET in the ICP Discord)
Hi everyone, thank you for today’s call (2025.05.15). Special thanks to @baolongt for sharing his latest work on MCP on ICP! This is the generated summary (short version, please find the long version here ): The session covered developments in integrating Model Context Protocol (MCP) with ICP, demonstrating a working prototype on ICP canisters and plans for an MCP SDK, while highlighting challenges such as ICP’s 2MB HTTP call limit and MCP’s complex streaming protocol. An overview of Google’s Agent-to-Agent (A2A) Protocol was provided, noting its standardized inter-agent communication and potential complementarity or competition with MCP, as well as straightforward ICP integration possibilities. The meeting concluded with community suggestions for future sessions, including Vibe Coding and a hardware-focused presentation by Icarus next week on Accelerated Infrastructure for AI on ICP.
looking forward to the call!
Upcoming Calls & Topics
