neuromem
A long-term memory engine that lets an AI agent keep who it is and what it has been through, and recall the memories that fit the moment by association.
The 32 public projects run by our three researchers. 5 personal projects are left off this list. The bars under each card count research journals per week since March 2026.
A long-term memory engine that lets an AI agent keep who it is and what it has been through, and recall the memories that fit the moment by association.
A memory layer that cuts forgotten context and hallucination in AI coding across sessions. It says "I don't know" when it doesn't, and answers with sources when it does.
A multi-agent runner that launches Claude Code, Codex and Gemini CLI unmodified as boxes on a canvas, wires them together, and shows who started when and what failed.
Runs the open-source memory server Honcho on our own machines as a long-term memory store shared by Claude, Codex and Gemini.
Gathers memory research from cognitive psychology and neuroscience to ground the design of a memory system modeled on human memory.
Draws the boundaries between our five memory projects and weighs on one scale what to keep and what to fold.
Feeds the same data to LLMs in seven formats, including JSON, YAML, CSV and Markdown tables, and traces why accuracy and token counts change.
Tested "does architecture help AI coding?" across 15 code structures and 209 cells. What separated cascading failures was not the structure but the model tier.
A benchmark that measures on several axes which model fits our own everyday use, rather than public leaderboards.
A browser-control layer with no LLM of its own. Any agent connects over MCP and handles web pages by meaning instead of screen coordinates.
Verifies by machine measurement, not human judgment, whether the path from Figma mockups to a design system and code stays faithful.
A CLI that scores the consistency of code, styles and design. Published on npm as dsmonitor.
A canvas where AI and people edit meaning together, not coordinates.
Runs image-generation workflows from Discord commands with n8n alone, without an always-on bot.
A macOS menu bar tool that shows usage across AI providers at a glance, and draws partial data without distorting what it means.
This lab's record system: researchers log their work as journals and refine conclusions into a knowledge graph. The numbers on this site come from it.
Surveyed how universities, companies and institutes keep research records, and set the format of our lab notes and weekly reports.
Builds an isolated Blender environment where AI edits and checks 3D-printing models (STL) directly, then redesigns real prints with it.
A handheld voice input device with six buttons and a microphone. Transcription runs on local Whisper on a Mac.
A local web tool for refining bike route drafts on a map and exporting them as GPX for a bike computer.
A macOS app that transcribes speech on the device, without sending audio to the cloud.
A local composing environment where AI writes and plays music, and people listen and edit in the browser.
Checks in Godot whether a deterministic puzzle game about opening blocked rail networks with the fewest works holds up.
A web app for whisky, wine and sake tasting notes and prices by shop, with taste analysis.
An experiment in writing a Korean web novel all the way to the end with AI.
A roguelite stock-survival game where you move the market with daily trading cards. Built with React Native and Expo.
Generates KakaoTalk themes from code instead of drawing them by hand. Change one color and everything regenerates.
Builds SDXL and ComfyUI image-generation presets that give the same result no matter who runs them.
A Chrome extension and Mac app that turns foreign text on screen into Korean in place.
A local voice assistant that adds only voice and a HUD on top of the Codex we already use.
A study project building a RAG chatbot that answers from uploaded documents.
Joined the "AX Talent War" hackathon by JocodingAX and Primer, and built Codex plugins that solve each company's problem.
No projects match these filters.
Numbers are counted from the research journal repository. Journals is the number of research journals written for a project; Cards is the number of finding, hypothesis and decision cards drawn from it.
Questions about our research, collaboration and requests for materials are all welcome by email.