April 7, 2026
Product Release
ClaudeCode Native Plugin: Native & Long-term Memory for ClaudeCode
Anthropic made a big change yesterday: starting April 4–5, 2026, Claude Pro and Max subscriptions no longer cover usage through third-party tools like OpenClaw. What used to feel like unlimited agentic workflows suddenly became pay-as-you-go or more expensive.
The community has been frustrated for the same reasons: broken workflows, surprise bills, and the sinking feeling that your entire agent setup could disappear overnight if the next policy change hits. If you’ve been building long-running agents, multi-day projects, or complex coding flows, this change disrupts your workflows.
But here’s the good news:
We built ByteRover - an open-source Karpathy-inspired memory system that that gives your agents permanent, shared memory across OpenClaw, Claude Code, Hermes, Cursor, and more.
It maintains and recalls the exact context even with lightweight models likes Claude Haiku.
The Real Pain Everyone Is Feeling Right Now
Workflows break or get expensive: Your OpenClaw agents still work, but the cost jumps because of the new billing rules.
Unexpected extra charges: Heavy users who ran smooth agentic flows now face separate pay-as-you-go fees.
Loss of control: It feels like Anthropic can change the rules anytime after you’ve built your entire setup around their models.
Memory keeps disappearing: When you switch tools or adjust workflows, your agents forget everything and you start from scratch.
The Open-source Long-term Memory that Connects your Agents
Today we shipped the native connector for Claude Code.
Your agents inside Claude Code instantly get full automatic memory:
Recall the right context before every turn
Smart curation (Automatic Memory Management) after every turn
Context automatically organized into file-based hierarchy that persists across sessions
Because ByteRover already works natively with OpenClaw, Hermes Agent, all the tools now share the same memory folder. Your agents stay consistent no matter which tool you’re using.
Why This Actually Works (and Saves You Money)
Your memory travels with you
Bring your entire context tree in minutes:
OpenClaw → Claude Code → Cursor → Hermes → whatever comes next.
It lives in plain files you own - not locked inside any company.
Markdown-native architecture validated by Karpathy
Everything is automatically stored in organized, human-readable Markdown files. No black-box vector embeddings. You can open, edit, search, or Git-version them exactly like normal code.
This is the clean, deterministic approach @karpathy has championed for reliable agent memory.
Real production-grade features
Automatic brv query (recall) before every LLM call
Automatic brv curate (knowledge managemnet) after every turn
Self-reinforcing importance scoring based on actual usage
Git-like versioning and easy team sharing
50-70% fewer tokens on average - works beautifully even on cheap models like Claude Haiku
The Bottom Line
ByteRover lets you keep using OpenClaw + Claude Code exactly how you want, without paying extra just to maintain memory.
Your agents finally remember across tools and across policy changes, no more surprise bills. Just reliable, portable memory that belongs to you.
Get started in 60 seconds:
Install ByteRover CLI:
Install the Claude Code plugin:
You can then verify the installation with: brv-claude-plugin doctor
Check detailed installation guide here: https://github.com/campfirein/brv-claude-plugin