You can use IWE as memory backend for Claude Code

[gimalay]

IWE now works as a memory backend for Claude Code. The iwe plugin stores what an agent learns across sessions as plain markdown in the repository's own IWE workspace, next to the project's other documents, committed and reviewed like any other file.

What shipped:

• Memory in the repo. Captured notes are ordinary IWE documents in your workspace. No separate memory directory, no second graph. They link to your existing docs and are found by the same iwe find queries, and they travel with every clone, so a teammate's first session starts with what the team already knows.
• Nothing is written unless you pick it. /iwe:distill reads a session with you, proposes items with a quote from the conversation as evidence, and writes only what you keep. Every write goes through iwe create --strict, so it is validated against the schemas your workspace binds before it lands.
• MEMORY.md is the switch and the policy. One document, prose you own: what is worth keeping, how documents in this store are written, how to search before writing. Without it the plugin does nothing at all. /iwe:init writes it with you; /iwe:reflect reorganizes the store with you later (index fields, areas and hubs, entity pages, merges).
• A session-start index from queries you own. At the start of every session the plugin injects a token-budgeted list of memory, titles and keys only, built from the injection queries in the policy's frontmatter. Deterministic, no model in the loop; the agent fetches a document only when the conversation turns to it.
• Not only Claude. The three skills install into Codex, Cursor and OpenCode with npx skills add iwe-org/skills, and any agent can read and write the same store through the CLI or npx -y @iwe-org/mcp.

The design and the reasoning behind it are in the blog post: https://iwe.md/blog/memory-that-lives-in-your-project/

Setup instructions are in the plugin repo: https://github.com/iwe-org/skills#iwe-memory

It is new. If it proposes the wrong things, or the right things in the wrong shape, or a schema check refuses something it should not, I would love to hear about it here.