June 3, 2026
AI finally remembers the backlash
Show HN: Mnemo – local-first AI memory layer for any LLM (Rust, SQLite,petgraph)
This app says it gives chatbots a memory — commenters say “so does everybody else”
TLDR: Mnemo is a new tool that gives chatbots a long-term memory by saving useful facts locally on your computer. Commenters were split: some liked the cross-tool promise, while others said AI memory tools are already everywhere and can actually make chatbot replies worse.
A new Show HN project called Mnemo is pitching itself as a way to stop chatbots from acting like they’ve just walked into the room. The idea is simple enough for non-experts: it watches your AI conversations, saves key people, places, projects, and facts into a local database on your own machine, then feeds the useful bits back into future chats. No cloud, no account drama, just a downloadable tool that promises fast “memory” for your favorite AI assistant.
But the real show was in the comments, where the community instantly turned into the Memory Layer Jury. One camp basically yawned and said: here we go again. Several people argued that this is becoming the new “everyone has a podcast” for AI builders — another week, another memory tool. One commenter even urged the creator to add a big “Why Mnemo?” section up top, because the field is already crowded and blurry.
Then came the sharper hot takes. A skeptic warned that stuffing old memories back into an AI chat often makes answers worse, not better — the tech equivalent of a friend who won’t stop bringing up embarrassing old stories. Another person dropped competitor links like receipts in a group chat, while others said the feature may eventually get swallowed by larger AI apps anyway. Still, there was real interest from people who want project-wide memory that works across tools, not locked into one chatbot. In other words: cool idea, but the crowd wants proof, not vibes.
Key Points
- •Mnemo is presented as a local-first memory layer that persists LLM-derived knowledge across conversations using a SQLite-backed knowledge graph.
- •The service ingests raw text, uses a configured LLM to extract entities and relationships, deduplicates entities, merges aliases, and updates an in-memory petgraph graph.
- •Retrieval uses a six-stage pipeline: full-text chunk search, entity name search, graph expansion, relation filtering, scoring and ranking, and context prompt assembly.
- •The project supports multiple model providers, including Ollama, OpenAI, Anthropic, and OpenAI-compatible APIs, and is distributed as a static binary with no cloud dependency.
- •The article documents Docker, Rust binary, and Python SDK usage, along with API endpoints for health checks, ingest, retrieval, search, entity and chunk management, wiping memory, and statistics.