Why I chose plain Markdown over a database for AI agent memory

说明与用途

AI 代理缺乏持久上下文，作者创建了基于 Markdown 文件的本地优先内存存储，代理可读取，人类可编辑，并通过 Git 追踪历史。

问题

Agents lacked persistent context, requiring repeated explanations each session.

结果

Agents now load context from Markdown files, reducing repetitive prompts and enabling human review of memory updates, while retaining version history via Git.

工作流

• Create a workspace folder with subfolders for profile, stakeholders, context, context‑updates, meetings, outputs, and tasks.
• Store each piece of memory as a Markdown file with YAML frontmatter.
• Configure agents to read specific files (e.g., CLAUDE.md) to load context.
• Record meetings and transcribe locally using Whisper or SenseVoice.
• Use an agent to compare transcripts with existing memory and propose changes.
• Review and accept or reject proposed changes before they are written to memory.
• Version control the workspace with Git to track and revert changes.
• Optionally build a dashboard that edits the workspace via temp files and renames.
• Use an optional SQLite index for search across files.
• Bind dashboard to 127.0.0.1 to isolate from cloud agents.

证据

Every time I opened a new Claude Code session, I spent the first few minutes explaining my work again.

来源

Agents already read files well.

来源

Git gives history for free.

来源

The agent does not write to memory directly. It proposes changes, such as a new deadline, a changed decision or what a colleague expects, with the evidence from the transcript. You accept or reject each one.

来源

Plain files are simple until two writers touch the same folder.

来源