Show HN: A Governed Context Vault for Claude Code and Cowork (AGPL CLI)

Every Monday morning, somewhere in your company, someone opens Claude — in a terminal, in Cowork, it doesn't matter which — and starts teaching it the business. Again.

Here's the schema. Here's the security protocol. Here's why we never call that service directly, and why the pricing page hasn't matched the sales deck since March. Twenty minutes of re-briefing later, Claude is useful. By Tuesday, none of it survived the session.

That re-briefing tax is the cost you can see. The expensive one is invisible: Claude fills any gap in what you told it with whatever pattern looks plausible, and hands back an answer that sounds just as confident whether it's citing your actual policy or improvising one that merely sounds like it. Multiply that across every team and every session, and the business slowly drifts from what it actually decided — one convincing, unverified answer at a time.

The ContextNest plugin for Claude closes that gap. It installs the same way and behaves the same way whether your team runs Claude Code from a terminal or works out of Cowork — Claude stops guessing and starts working from a governed, version-controlled source of truth: retrieving from it, writing back to it, and flagging what's gone stale, without anyone leaving the conversation.

Search Finds Everything. It Doesn't Know What's Approved.

Enterprise search tools index everything — every Slack thread, every abandoned draft, every doc someone meant to delete. Indexed and authorized aren't the same thing. Point Claude at a flat folder of files and it can't tell your current security protocol from the one you retired eighteen months ago. Ask it a technical question and it will cite either one with identical confidence.

For developers: this is the difference between a folder Claude can search and a graph Claude can reason over. Search returns what matches the query. A governed vault returns what your org has actually approved, with a version history attached.

One Change Ripples Through Everything. You Steer Every Edit.

Knowledge doesn't change one document at a time. Rewrite a pricing tier and the sales deck goes stale. Update an API contract and three integration guides now describe endpoints that no longer exist. Revise a data-retention policy and the security runbook still points to the old rule. Change your positioning and half the site keeps repeating the line you just retired. In a folder of files, none of that connects — each document sits alone until someone stumbles onto the mismatch.

Because ContextNest links your knowledge as a graph, Claude sees those connections while it works. And instead of picking a side on its own, it surfaces the conflict and waits. Claude doesn't decide which version is correct. It flags what it found, applies the version you've already approved to the work in front of it, and holds everything else for you to resolve. The AI doesn't drive the change. You do.

This happened to us last week. While rebuilding a page, Claude surfaced a conflict between our messaging architecture doc and a newer ruling from a partner review, about how we describe our own technology — two documents, both marked authoritative, saying opposite things. Instead of picking one, Claude applied the newer ruling to the work in progress and queued the older document for me to reconcile.

Claude, mid-task