Shared memory for AI agents

I use a shared personal memory for Claude and Codex that I can open and edit on my iPhone.

My Obsidian vault holds context about me: my background, preferences, goals and how I like to work.

CLAUDE.md and AGENTS.md point the agents to that shared knowledge. Both Claude and Codex can read the relevant notes and write updates back, so the context can be maintained across tools.

I chose Obsidian for a few practical reasons:

• The knowledge lives in Markdown files I control, independently of any one AI provider. I choose where those files are stored. They could stay on my computer, or I could use my own NAS for storage and backups. Cloud storage is optional.

• I can read, edit and correct the notes myself. In my current setup, iCloud Drive syncs the vault across my Mac, iPhone and other Apple devices, so I can work on it wherever I am.

• The files can be version-controlled with Git. I can track changes, compare revisions and restore an earlier version when needed.

• I also have Claude Code integrated into Obsidian. I can ask questions about the material and work with an agent to edit it directly inside the app.

I’ve also set up recurring routines across my agent tools to refresh their memory from the vault and bring relevant updates back into Obsidian. That helps keep the shared notes and the agents’ own context aligned.

iCloud handles file synchronization between my devices. The agent routines handle the context updates. What a cloud model processes still depends on the information I send to it.

Keeping this useful takes some discipline: confirmed facts need to be distinguished from suggestions, changing information needs a date, and contradictions need to stay visible until they are resolved.

A workspace such as Notion or a plain Markdown repository could also be a starting point. Obsidian gives me an interface I use myself, with agents working on the same underlying notes.

How are you keeping context consistent across the AI agents you use?