ZeroDB Memory: Give Claude Code Persistent Memory Across Sessions

# ZeroDB Memory: Give Claude Code Persistent Memory Across Sessions Every Claude Code session starts the same way — with Claude knowing absolutely nothing about your project. You re-explain the same architecture. You correct the same mistakes. You paste the same context. Every. Single. Session. We built ZeroDB Memory to fix that. ## The Problem: Context Loss Is Real Friction If you use Claude Code daily, you already know this pain. You open a new terminal, run `claude`, and immediately you're back to square one. The codebase has no memory of what you talked about yesterday — which module is owned by which engineer, what that cryptic environment variable does, why you use PgBouncer on port 6432 and never connect to Postgres directly. You might have a `CLAUDE.md` file at the root of your repo. It helps — but keeping it current is a manual chore. It tells Claude what you wrote down at some point in the past, not what you and Claude figured out together last Tuesday. And every time you onboard a new project or spin up a fresh worktree, the clock resets. WozCode and similar tools help by compressing tokens during a session — great for keeping your context window lean while you're working. But token savings within a session don't help when the session itself ends and the slate wipes clean. The problem isn't session length. It's session continuity. The result: developers waste the first 5–10 minutes of every Claude Code session re-establishing the same shared context. For a team of five engineers running two sessions a day, that's hundreds of hours of lost productivity per year — not counting the bugs caused by Claude giving advice that ignores conventions your team has already established. ## Introducing ZeroDB Memory for Claude Code ZeroDB Memory is an open-source Claude Code plugin that gives Claude a persistent, project-scoped memory layer. It installs in one command, requires no changes to your workflow, and starts working immediately. **At session end**, a Stop hook fires and Claude extracts key facts from the conversation — architecture decisions, conventions discovered, bugs fixed, ownership rules, environment quirks — and stores them in your ZeroDB project using semantic embeddings. Not raw transcripts: structured, searchable memories. **At session start**, a PreToolUse hook fires, loads memories tagged to the current git repository, and injects the most relevant ones into Claude's context — ranked by semantic similarity to whatever you're about to work on. Claude walks into the session already knowing your stack. Beyond the automatic flow, you get four slash commands for manual control: `/remember`, `/recall`, `/forget`, and `/memory`. Use them to pin critical facts, query what's been stored, or clean up outdated information. ## How It Works ### Install the plugin ```shell /plugin install github:AINative-Studio/zerodb-claude-plugin ``` ### Set your API key ```bash export ZERODB_API_KEY=your_key_here ``` Get your key at [ainative.studio/signup?ref=claude-plugin](https://ainative.studio/signup?ref=claude-plugin) — 500 free credits included. ### Pin memories manually Use `/remember` to capture anything important mid-session: ``` /remember The payments module is owned by Sarah Chen — never refactor without her sign-off /remember We use PgBouncer on port 6432, never connect to 5432 directly /remember Railway internal DNS doesn't work from Kong — always use public URLs /remember The Celery beat scheduler runs on a separate Railway service, not the main backend ``` These get stored with semantic embeddings so they surface contextually, not just as a flat list. ### Query memories on demand ``` /recall how do we connect to the database → Found 2 memories: [convention] Database connections go through PgBouncer on port 6432, not 5432 [architecture] Railway internal DNS unreachable from Kong — use public URLs ``` The recall uses vector similarity search — so asking "how does auth work" finds the memory about JWT tokens even if that memory doesn't use the exact word "auth". ### The automatic flow When you end a session, the Stop hook sends the conversation to ZeroDB Memory, which extracts structured memories and stores them tagged to your current repo (by git remote URL). Next time you open Claude Code in that repo, the PreToolUse hook pulls the most relevant memories and injects them before Claude touches a single file. You don't have to do anything. Claude just... remembers. ## The CLAUDE.md Generator One of the most useful features for teams is `/memory export --format claudemd`. After a few weeks of working with ZeroDB Memory, your accumulated memories represent the real, lived knowledge of your codebase — not the idealized documentation someone wrote at project kickoff. Run this command and the plugin generates a `CLAUDE.md` draft from your stored memories, formatted as Claude Code conventions. It's the difference between writing documentation and having it emerge organically from how your team actually works. ## ZeroDB Memory vs. WozCode We get this question a lot, so let's be direct: they solve different problems and work better together than either does alone. **WozCode** reduces token consumption during a session by compressing file reads and context. It makes individual sessions cheaper and faster — real value if you're running long coding sessions. **ZeroDB Memory** solves cross-session continuity. It doesn't save tokens in the current session; it prevents you from wasting the first 10 minutes of every session re-establishing context that Claude already discovered last time. If you use both: WozCode keeps your sessions lean, ZeroDB Memory makes each new session as informed as if it were a continuation of the last. That's the combination we'd recommend for any developer using Claude Code more than a few hours a week. ## Free to Start ZeroDB Memory is free to install and open source. Memory operations bill to your ZeroDB account: - **500 free credits** on signup — roughly 20–25 sessions worth - **$9/month** for 5,000 credits — about 200 sessions, or a full month of daily use - **$29/month** for 25,000 credits — team tier, covers multiple developers Sign up at [ainative.studio/signup?ref=claude-plugin](https://ainative.studio/signup?ref=claude-plugin). ## What's Next: Team Shared Memory Phase 4, coming in May, brings **shared team memory**. Right now, memories are per-developer. In Phase 4, you'll be able to connect your whole engineering team to a single ZeroDB project — so when Sarah learns something about the payments module and stores it, every other developer on the team benefits from that knowledge in their next Claude Code session. Think of it as a living, semantic knowledge base for your codebase that grows automatically as your team works, without anyone having to write documentation. --- **Install today:** [github.com/AINative-Studio/zerodb-claude-plugin](https://github.com/AINative-Studio/zerodb-claude-plugin) **Docs:** [docs.ainative.studio/guides/claude-plugin](https://docs.ainative.studio/guides/claude-plugin) **Free signup:** [ainative.studio/signup?ref=claude-plugin](https://ainative.studio/signup?ref=claude-plugin)

ZeroDB Memory: Give Claude Code Persistent Memory Across Sessions

Check your site's AX Score