Clude launches memory agent
Clude launched on Product Hunt with an agent memory platform claiming an 81.3% LongMemEval score and 1.96% hallucination rate, plus support for npm/Python/MCP frameworks and on‑chain verification. The product is explicitly built in public and targets long‑term memory/recall use cases for agents. (x.com)
Clude commits each stored memory as a SHA‑256 hash and publishes those commitments to Solana using an on‑chain memo/Anchor pipeline, with the project documenting the approach on its site. (clude.io) The system organizes memory into four typed layers — episodic, semantic, procedural, and self‑model — and runs automated “dream” consolidation and decay cycles to convert episodic traces into longer‑lived knowledge. (colosseum.com) (clude.io/docs.html) Clude is distributed as a self‑hosted Model Context Protocol (MCP) server plus SDKs and advertises integrations with Claude Desktop, Claude Code, Cursor and connectors that route any model (Claude, GPT, Llama, Mistral) through a Venice AI connector. (mcpmarket.com) (clude.io) The project says memories are encrypted client‑side with user keys, supports exportable “memory packs” for portability, and publishes an MIT‑licensed SDK on GitHub for self‑hosting and local deployment. (clude.io) (github.com) A live demo dashboard shows real‑time brain metrics — the demo snapshot lists about 1,000 total memories and 994 of them committed on‑chain — illustrating their end‑to‑end hashing and commit flow. (clude.io/demo.html) Clude is also tied to a Solana token narrative (token CLUDE) with DEX listings and market coverage; exchanges have announced listings and third‑party writeups track early trading activity while marketplaces index the on‑chain contract. (toobit.com) (blog.tapbit.com) (opensea.io)
