Persistent memory & version control for Claude Code.
It doesn’t just retrieve — it thinks before it answers. 100% recall where similarity search guesses, and it never hands back anything you didn’t store.
Contextual analysis picks the memory that fits what you actually asked — 100% where plain vector search can’t.
When the answer isn’t there, it says so — instead of hallucinating a confident wrong one the way embedding memory does.
It returns your stored data, not a paraphrase — so 100% of what comes back is real. If it doesn’t, it wasn’t there.
A single executable for Windows 10 & 11.
O-> ?( name("background") ) OK: 97 match(es) background-001.html background-002.html background-003.html … 97 files — direct-addressed, no crawl, no index build O-> grep lineDiff screensaver-codescrub.html: function lineDiff(aText,bText){ screensaver-codescrub.html: var ops=lineDiff(prevText,text), … 2 matches · searched 123 files: 123 store-cached, 0 disk O-> history screensaver-codescrub.html LOG screensaver-codescrub.html (19 versions) v1 2026-07-25 20:25:32 8174b v17 2026-07-26 01:50:31 14698b v19 2026-07-26 01:54:49 14939b every save kept — roll back to any one
trying to break it and failing
Real queries against a live store — not a mockup. A grep that never touches the disk, every version of every file kept. Copy them, run them, try to make it miss.
Persistent memory and source-code version control
built for Claude Code.
Durable, context-efficient memory. Claude offloads what it learns to a store and pulls it back on demand — so it boots up already knowing your project, and your context window (and token bill) stays lean.
The space between your git commits. Every save is backed up automatically — roll back to any moment, not just the ones you remembered to commit.
Coordinating fleets of Claude Code agents from one surface — task queues, heartbeats, and live channels instead of a pile of terminal windows. On the roadmap, not shipping yet.
You commit your milestones to git. But between those commits are hours of work: experiments, half-finished refactors, the version that worked five minutes ago. 0verload versions every single save automatically, so every one of those in-between states is recoverable.
Keep checking your major versions into GitHub. 0verload keeps everything in between — every save you'll wish you had a backup of, now you do.
Every turn re-reads your whole context window — the bigger it gets, the more each turn costs and the slower it runs. 0verload's memory flips that: Claude keeps only pointers hot and offloads the detail to a durable store, pulling facts back exactly when needed.
Effective memory goes unbounded while your working context — and your token spend — stays small. Start using the memory and your budget goes further.
0verload speaks MCP natively, so its tools show up in Claude Code the moment you install it: versioned file reads and writes, full history, and memory — all callable directly. It auto-enrolls the project you're working in and persists your memory across sessions.
It also speaks A2A and OpenAI-compatible APIs, so it drops into whatever else you build with, too.
Four qwen‑7B bots with an 8k context window — the last thing you’d point at a million‑token benchmark — answering non‑stop against one shared 0verload store. 0 questions answered so far and counting. Every number here updates live.
This is the tiny‑model floor: the accuracy is a small 7B struggling to reason, and that’s the point — the recall column (the store finding the fact) is 0verload doing its job in milliseconds, every time. See the full live benchmark, every unabridged result →
Cancel anytime. Subscriptions are billed and secured by Stripe.