Top AI Product

Agent memory is the hottest infra race of 2026, and Tencent Cloud just walked in with an open-source, fully local entry. TencentDB-Agent-Memory is an MIT-licensed long-term memory system for AI agents: 8,283 GitHub stars, +123 today, still on the Trending board.

Two memories, zero external APIs

Short-term memory compresses bulky tool logs into compact Mermaid symbols instead of dumping raw text back into context. Long-term memory runs a 4-tier pipeline: L0 raw conversations → L1 atomic facts → L2 scenario blocks → L3 user persona. Everything sits on local SQLite with sqlite-vec — no external API calls, your data never leaves the machine. That’s the real differentiator: most memory products are flat vector piles in someone else’s cloud; this is structured distillation on your own disk.

The numbers sell themselves. Plugged into OpenClaw, token usage drops up to 61.38%, task pass rate rises 51.52% relative, and PersonaMem accuracy jumps from 48% to 76%.

For builders: TypeScript SDK, one npm plugin

It ships as a TypeScript SDK with an npm plugin for OpenClaw and a Docker image for Hermes, using hybrid BM25 + vector retrieval. Typical use case: a coding agent that still remembers your stack preferences three weeks later — without replaying the whole history.

You Might Also Like

• Openviking Treats ai Agent Memory Like a File System and 9k Github Stars say its Working
• Deepseek tui Tops Github Trending a Claude Code Clone Wired to Deepseeks api
• Heretic Just hit Github Trending and the ai World has Opinions
• Pentagi Just hit 1 on Github Trending and Yeah its Worth the Hype
• Pageindex Just hit Github Trending and it Might Make you Rethink rag Entirely