How to Give Claude Code Persistent Project Memory
Ville YlläsjärviFounder of RAAVA strong Claude Code session can still leave the next session guessing. Here is a practical way to keep product decisions, task ownership, and verification outside the chat transcript.
The session amnesia problem
Claude Code can ship a feature in one session and forget the product context in the next. Founders re-explain the app. Agents re-discover the same files. Verification status lives in chat scrollback.
A rules file helps with stable instructions. The missing piece is live state: the task in progress, the branch that owns it, the last verification result, and the decisions waiting for a person.
What to store in shared memory
At minimum: product brief, active goals, open requirements, prioritized tickets, branch lanes, and claims. When an agent runs `raav submit`, record files touched, summary, verification command, and risks.
Keep founder-confirmed facts separate from the agent's working assumptions. That one distinction prevents a surprising amount of false certainty.
Setup in one repo
Install the RAAV CLI, run `raav init`, and paste the onboarding prompt from the install page into Claude Code. Seed product memory with onboard-and-plan, then pull scoped work with `raav next`.
The install guide has the exact prompt and checks. RAAV keeps this record locally and does not call a model to maintain it. Founder testing is currently invite-only.
When memory beats another prompt
Use RAAV when multiple sessions touch the same launch, when two agents might collide on files, or when a founder needs an audit trail for handoffs and investors.
For PM onboarding without reading every file, pair memory with the codebase onboarding use case.
Test RAAV on a real repo
Request access to a controlled founder cohort. Joining creates no subscription or charge.
Join waitlist