Two Kinds of Memory

A developer introduced virage, a new tool that creates a searchable vector index of a codebase, complementing the previously released kms tool for capturing curated decisions. Virage uses AST-aware chunking and semantic search to answer 'where' and 'what' questions about code, while kms handles 'why' questions with citations. The developer emphasized that the two tools address distinct memory problems and are not integrated.

I asked my AI agent where the retry logic lived. Four seconds later: three files, one shared helper, a comment linking to a Slack thread that doesn't exist anymore. Found it cleanly. Then I asked why the retry count was capped at three instead of five. Silence, then a guess. It could find the code. It had no idea why the code said what it said. Two failures wearing the same complaint — "the AI doesn't know this project." One's a search problem. The other's a memory problem. I'd already fixed the memory one; that's what this whole series has been about. The search problem needed a genuinely different tool. kms only knows what someone deliberately wrote down as a decision, fact, or guardrail. Curated, citable, checkable — query answers "why does this work this way" with a real citation, because the answer is sitting in a file someone wrote on purpose. Most of a codebase isn't a decision, though. It's just code — functions, old comments, a README paragraph, a PR nobody formalized because it never rose to that level. kms has nothing to say about any of it. Not broken. Never the job. virage is a separate vivantel tool for exactly that gap — a git-aware pipeline that turns a codebase and its docs into a searchable vector index. virage init virage index virage query "how does authentication work?" init writes virage.config.json via an interactive wizard. index chunks the repo — AST-aware, so a function stays one semantic unit instead of getting sliced at an arbitrary line count — and embeds it. query searches by meaning, not keyword: ask about authentication and it finds the relevant code even if that word never appears in it — something grep structurally can't do. Every stage is swappable: an OpenAI-compatible embedder or a fully local ONNX model, an embedded file-based vector store for a solo repo or Postgres/Qdrant for a team, an optional reranker on top. Same CLI regardless of what's behind it. "AI memory" tools promising a single system for this exist — Mem0, Letta, Cognee, and similar. They solve a third, different problem: raw recall of accumulated facts or conversation history, closer to virage's mechanism than kms's, but aimed at remembering what happened, not retrieving a curated decision a human stood behind. None separate "a fact" from "a commitment the team made" the way kms does, and none are built for a git repo's own structure the way virage is. An earlier piece in this project's kms series described a real mess: auth split across three places, nobody could say why. virage index finds all three in seconds — every file touching auth, ranked by relevance, whether or not "auth" shows up in a variable name. Search problem, solved. It can't tell you why it's split three ways. That answer, if it exists, lives in a decision someone wrote down — the kind bootstrap 's gap-fill mode goes looking for, or query retrieves with a citation once it's captured. virage has no opinion on that question. Never asked one. Neither tool is weaker — they answer different questions, where is this and virage query for "where"/"what" questions. kms query for "why" questions. They're two separate tools today, not one integrated system — running virage doesn't feed kms 's decisions, and vice versa. Worth saying plainly rather than implying a seamlessness that isn't there. What they share is a practice, not a product: don't let either kind of memory live only in someone's head. If this is useful, a star helps other developers find both: ⭐ Star vivantel/kms https://github.com/vivantel/kms · ⭐ Star vivantel/virage https://github.com/vivantel/virage /plugin marketplace add vivantel/kms /plugin install kms That's Claude Code. Codex reads the same skill set through its own plugin manifest. Kilo Code CLI reads it too, once kilo.jsonc points at the published skills manifest. npm install -g @vivantel/virage virage init Claude Code again first, via virage's own agent plugin. GitHub Copilot reads the same index through its own agent plugin. Codex reads it through its own agent plugin. Antigravity reads it through its own integration. If you already run kms, virage index on the same repo takes about a minute and costs nothing to try. If you run neither yet, start with whichever question you actually have: something you can't find, or something you can't explain.