Continual Learning with .md

Name: Continual Learning with .md
Availability: InStock
Author: wenhan_zhou

by wenhan_zhou·Apr 13, 2026·34 points·34 comments

Visit Project View on HN

AI Analysis

●MidNiche GemBold Bet

Scrappy v1 with dangerous setup flags and zero stars on GitHub.

Strengths

•Hierarchical date-based organization mirrors how humans recall temporal context.
•Shell command retrieval means no vector database dependencies required.

Weaknesses

•Requires --dangerously-bypass-approvals flag which is a non-starter for most.
•One commit, zero stars, no evidence this works better than established RAG.

Post Description

I have a proposal that addresses long-term memory problems for LLMs when new data arrives continuously (cheaply!). The program involves no code, but two Markdown files.

For retrieval, there is a semantic filesystem that makes it easy for LLMs to search using shell commands.

It is currently a scrappy v1, but it works better than anything I have tried.

Curious for any feedback!

Similar Projects

AI/ML●●Solid

A-Life Powered by LLMs

Filesystem-dwelling AI agents with bio-feedback drives in a crowded agent space.

Rabbit HoleBold BetNiche Gem

DeanoC

203mo ago

AI/ML●●Solid

Agent-fs – The missing filesystem for agents (and humans)

MCP integration lets Claude Code use this as long-term agent memory.

Big BrainNiche Gem

tarasyarema

203mo ago

AI/ML●●Solid

Letta – Git-Based Memory for Coding Agents

Context Repositories — treating agent memory like a git repo you can diff, branch, and version — is a clever, developer-friendly twist on long-term LLM state. Letta Code’s persisted, model‑agnostic agents and the Conversations API make the product feel like a coherent stack for production agents, though the trick will be real-world scale, merge semantics, and cost of storing rich context over time.

Big BrainBold BetSlick

morawr

204mo ago

Developer Tools●●●Banger