[FEATURE] Persistent project memory for claude.ai — like CLAUDE.md, but for the browser

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Every claude.ai session starts from zero.

I run several SaaS products simultaneously. Every time I open claude.ai I re-explain the same things: which project, the stack, the deployment, decisions made last week, what already works and what failed. The memory feature helps with personal preferences — but it's not structured project memory.

Claude Code solves this elegantly: CLAUDE.md + skills load at session start. The model knows the project before my first message. There's no equivalent for claude.ai.

I built an open-source skill, Project Brain (https://github.com/OoneBreath/claude-code-project-brain), that gives Claude Code a navigable map of projects — stack, decisions, pitfalls, what's done and verified. 118 clones in 14 days with zero promotion. The demand is there.

But I can't use it when talking to Claude in the browser. The irony writes itself.

Proposed Solution

• A persistent context file attached to a claude.ai account/project, loaded automatically at session start — CLAUDE.md for the browser
• Or simply: "pin a file to this project" that Claude reads before every conversation

This would benefit every power user with ongoing complex work — developers, founders, researchers.

Alternative Solutions

Currently I work around this by manually pasting context at the start of every conversation, or relying on the memory feature (which captures preferences, not structured project state). I also built an open-source skill for Claude Code (Project Brain) that solves this on the CLI side — but there's no way to use it in the browser.

Priority

Medium - Would be very helpful

Feature Category

Other

Use Case Example

1. Monday: I work with claude.ai on my SaaS product — explain stack (FastAPI, Postgres, OVH), deployment, recent decisions.
2. Tuesday: new session — Claude knows nothing. I re-explain everything (10-15 min, thousands of tokens).
3. With this feature: a pinned context file loads at session start. Claude knows the project before my first message.
4. This saves 10+ minutes per session and prevents Claude from mixing up details between my different projects.

Additional Context

Reference: https://github.com/OoneBreath/claude-code-project-brain — 118 clones in 14 days with zero promotion shows the demand for structured project memory. Does anyone else feel this gap?