RepoOps / Memory

Blog

Field notes on attribution, AI security, remediation, LLM cost metrics, and repository memory.

  • Every security control in RepoOps, and the code behind it

    A full tour of the security architecture across the local app and the hosted dashboard. Redaction, write-trust, signed packs, the enforcement gate, the kill-switch, the credential broker, tenant isolation, and the limits we will not paper over. Every claim names the module that implements it.

    · RepoOps Team

  • Eval engineering, and what we shipped

    We said eval engineering was the next rung. Here is the six-PR program that built it into the accountability loop, the outside validation, and the caveats we are keeping honest about.

    · RepoOps Team

  • The ten memory systems, and the one thing the list forgets

    The AI field keeps naming new kinds of memory: short-term, long-term, semantic, episodic, vector, graph, and more. RepoOps runs all ten. Here is the map, and the one thing every version of the list leaves out.

    · RepoOps Team

  • Prove it

    AI coding runs on unverified claims. This week we shipped the tools that check them, and pointed the first one at our own marketing.

    · RepoOps Team

  • After graph, the ladder keeps climbing: eval, memory, trust

    Prompt to context to harness to loop to graph. Each rung makes a bigger unit programmable. Here are the next three, and the three things every rung so far has left out.

    · RepoOps Team

  • Building a world-class daemon

    The capture daemon is the quiet foundation of RepoOps: it is what makes your proof-of-work and cost visibility always current. We audited it against the best daemons in the world across supervision, security, performance, fault tolerance, and updates, found where we were already strong and where the gaps were, and closed them. Here is the story, honestly told.

    · RepoOps Team

  • Who audits your AI bill?

    A reported $16.6 million invoice to a free-plan user. An auditing startup says it found $1.7 million in overcharges across $34 million of AI invoices. Metered AI billing is complex and mostly unverified. Here is how independent reconciliation works, and what RepoOps is building.

    · RepoOps Team

  • Every AI has memory now. None of it is yours.

    ChatGPT, Claude, and Gemini all remember you, inside their own walls. My Brain is the person-owned, portable brain that follows you across every AI tool, your repos, and your life. You own it, you read it from anywhere, and it reasons over what you actually know.

    · RepoOps Team

  • Why we built RepoOps

    The launch essay. Why "your AI wrote the code, RepoOps remembers it" needed a product, not a blog post.

    · RepoOps Team

  • Markdown doesn't rot. Unverified markdown does.

    A sharp argument is going around that AI agents built on piles of prompts and docs fall apart the moment production changes. It is right about the disease. We think it is slightly wrong about the cure, and we just fixed a textbook case of it in our own codebase.

    · RepoOps Team

  • AI code is now 27% of production. Review hasn't scaled to match.

    One line in four is already AI-authored, and the number grew five points in a quarter. Review bandwidth hasn't moved. The accountability gap is the structural problem RepoOps is built for.

    · RepoOps Team

  • The loop you can actually afford

    A popular post says stop prompting and start designing loops. It is mostly right. The part it leaves out is the bill, and who is paying it.

    · RepoOps Team