I'd reason through tradeoffs, settle on an architecture, make a dozen small decisions. Then the session ends. Next session has no idea what happened.

CLAUDE.md helps, but it's static and loads into every message. Most context only needs to be there when it's relevant. The hardest stuff to capture is the reasoning. I'll spend 30 minutes thinking through a plan with Claude, land on a direction, and none of it survives to the next session.

wire-memory is an open source Claude Code plugin that gives your agent persistent memory. It captures decisions, corrections, patterns, and preferences as you work. Before responding, the agent searches past context for anything relevant. No extra prompting from you.

Memory is scoped by project and user. Project scope means teammates can share the same memory. One person's corrections are available to everyone in real time, outside git.

No account required. You get a free ephemeral container for 7 days.

Under the hood, the plugin is thin on purpose. Everything goes into a Wire container. Wire handles storage, relationship discovery, structured and unstructured data, and more. Your agent gets 5 MCP tools to search, write, explore, delete, and analyze its own memory. Same container works from any MCP client.

Transcript capture is opt-in. Secrets are auto-redacted.

Data lives on Cloudflare infra so it's available at edge, making retrieval quick. Ephemeral container expires after 7 days but a plugin command lets you create an account to keep it.

Plugin is MIT, Wire itself is closed source. It's my startup. Happy to answer questions.