Lore: A Local LLM Agent That Remembers Everything You Tell It
The friction problem with note-taking apps isn’t organization — it’s capture. By the time you’ve opened the app, found the right folder, and decided on a title, the thought is either gone or you’ve talked yourself out of writing it down.
Lore solves this with a single design decision: a global hotkey (Ctrl+Shift+Space on Windows/Linux, Cmd+Shift+Space on Mac) that pops up a chat bar from the system tray. Type the thought. Done. Everything else — classification, storage, embedding, retrieval — happens automatically in the background.
What It Actually Does
Lore runs entirely on your machine via Ollama (local LLM) and LanceDB (local vector database). When you type something, it classifies your input automatically into one of four types:
- Thought — stores it with a timestamp and embedding
- Question — queries your stored thoughts via RAG and answers from your own notes
- Command — updates or manages existing entries (“mark the milk task complete”)
- Instruction — permanently changes how Lore responds to you
That last one is the most interesting. You can tell Lore:
“From now on, when I ask for my to-do list, show items as bullets with an emoji for each.” “Always end a response by listing the source rows and dates that helped you answer.”
It remembers these as persistent behavioral instructions. That’s not note-taking — that’s agent customization, without writing a single line of config.
The Recall Side
Natural language retrieval over your own stored thoughts:
- “What did I note at standup today?”
- “I’m about to leave — anything I need to do on the way home?”
- “What was the Stripe webhook event that caused the refund bug?”
The last example is from the README and it’s the most telling. Developers already do this informally — copying a curl command into a Slack DM to yourself, pasting a stack trace into a note “in case it comes up again.” Lore is that habit, but with a retrieval layer that actually works.
Where It Sits in the Personal AI Memory Landscape
The personal memory space is getting crowded, and it’s worth being clear about where different tools fit.
Lore is consumer-end: a desktop app, no code required, works out of the box after installing Ollama. The target user is anyone who wants a private second brain without setting up infrastructure. It captures your thoughts — the things you explicitly tell it.
soul.py (our own project) solves a different problem: giving AI agents persistent memory across sessions. When you close a conversation with an LLM and start a new one, it remembers nothing. soul.py is the layer that fixes that — it embeds conversation history, decisions, and learned context so the agent carries continuity across sessions. The user doesn’t directly interact with soul.py; the agent does. It’s infrastructure for developers building persistent AI assistants, not a personal note-taking tool.
The distinction: Lore stores your thoughts for you to query. soul.py stores an agent’s context so the agent can be consistent across time.
They’re complementary. An agent running soul.py for session continuity could in principle also query a Lore-style database of the user’s personal notes for richer context — that’s actually closer to how human memory works than either system alone.
The Honest Caveats
Lore is early. The roadmap lists “improved agent and software reliability” as the next milestone — which is a polite way of saying the current version is a proof of concept. Multi-device sync isn’t there yet, which is a real limitation for anyone who switches between machines.
The Electron stack also means it’s heavier than a native app. If you’re already RAM-constrained running Ollama locally, adding an Electron tray app adds to the load.
MCP support is on the roadmap. When that lands, Lore becomes connectable to other tools — agents could query your personal Lore database as a tool call. That’s the version that gets genuinely interesting for agentic workflows.
Setup
Download from the Releases page — Windows .exe, macOS .dmg, Linux .AppImage. Install Ollama if you haven’t already, then open Lore settings from the tray icon, navigate to Models, and pull an embedding model and an LLM. Lore recommends models based on your hardware specs.
That’s it. Press the hotkey and start capturing.