Here's where things are at with @cludeproject today 1. The memory engine. Agents store and recall context across sessions reliably. 1.3M memories, 600+ agents using it. The core API is stable and handles production traffic. 2. Retrieval quality. We went from 55% to 81.3% on LongMemEval after rebuilding the search pipeline. That puts us competitive with funded teams in the space. Hallucination rate sits at 1.96%. 3. Developer experience. npm and Python SDKs are out, MCP works, you can plug Clude into an agent in a few lines. 4. On-chain verification. Every memory gets a hash on Solana. Verifiable, tamper-proof, we store the hashes of the memory and do not display the content of the memory outright. 5. Leveraged on Venice infra to bring privacy first models to users. What's still rough: 1. The frontend. Dashboard has been unstable. Auth flow, wallet scoping, deploy issues. It's been the most frustrating part of the build. Getting rebuilt properly. 2. Improving our speed to recall; recent enhancements brought up our retrieval accuracy and reduced hallucinations but at the cost of speed. Trying to stretch this piece of the puzzle. Jumping on a stream tomorrow 🙏