You bought the Knowledge Vault plugin, you already have Claude Code running, and now you're looking at a README that reads like it was written for the person who built it. This is the other version — the one that assumes you've never installed a plugin before and walks the whole path with you, from two slash commands to asking your first question. The steps are short; what makes them feel daunting is that they're unfamiliar, and unfamiliar is what a walk-through is for.
Set aside about ten minutes. One step — getting an OpenAI key — sends you to a different website for a minute, so it helps to have a browser tab handy. Everything else happens inside Claude Code, where you already live.
01/What you'll get
Think of the plugin as a librarian for everything you've already told Claude Code. It reads through your material, files each piece on the right shelf, and when you ask a question it walks straight to the page that answers it — instead of you scrolling back through weeks of conversation trying to remember where a decision got made.
It reads three things, all of them already on your machine:
- Your past conversations — the back-and-forth from earlier Claude Code sessions in this project.
- Your saved notes and documents — the reference material you've collected for the project.
- What the assistant remembers — the running memory it keeps about your preferences and past decisions.
Once it has read all three, a question you'd normally have to go hunting for becomes a one-liner:
- You ask: "What did we decide about the pricing tiers last month?"
- It answers with the exact passage where you settled it, and tells you which conversation it came from — so you can trust the answer instead of taking its word for it.
02/Before you begin
Two things need to be true before the install will work. Sorting them out now saves you a confusing failure halfway through.
- Accept the GitHub invite. The plugin lives in a private repository, and access is granted by invitation after your purchase. Until you accept that invite, your GitHub account can't see the plugin at all — so the install will stop partway with no obvious reason why. Check your email for the invite, or look under your GitHub notifications, and accept it before you start.
- Have an OpenAI account ready. The plugin uses OpenAI to read and file your material, and it uses your account to do it — which means you stay in control of the cost and the data. Step 2 below walks you through getting the key; the OpenAI API keys page is where you'll do it. I'm linking out rather than showing screenshots because that page changes its layout often enough that a screenshot would lie to you within a month.
03/Set it up in six steps
Six steps, in order. Each one tells you what's about to happen before it happens, so you're never typing a command and hoping. Run them top to bottom and don't skip ahead — a couple of the later steps depend on the earlier ones having finished.
Step 1 — Install the plugin
Two commands, typed straight into Claude Code. The first one registers the Tenkr marketplace so Claude Code knows where to look. The second one pulls the plugin down and installs it — and this is the command that reaches into that private repository. If you skipped accepting the GitHub invite, this is exactly where it stops, because your account still can't see the repo. The @tenkr-plugin-marketplace on the end is required; it tells Claude Code which marketplace the plugin comes from.
// 1. Register the Tenkr marketplace /plugin marketplace add TenkrAS/tenkr-plugin-marketplace // 2. Install the plugin (the @marketplace suffix is required) /plugin install knowledge-vault@tenkr-plugin-marketplace
Step 2 — Get your OpenAI key
This is the one step that leaves Claude Code, and it's the only fiddly part of the whole process. You're fetching a key now because the setup command two steps from here is going to ask you to paste one — and you don't want to be scrambling for it mid-setup. Do it in this order:
- Open the OpenAI API keys page and sign in.
- Create a new secret key. Give it a name you'll recognise later, like
knowledge-vault. - Copy the key the moment it appears. OpenAI shows you the full key exactly once and never again — if you lose it, you make a new one.
- Confirm your OpenAI account has billing set up. A key with no billing behind it can't spend anything, so the plugin would have nothing to work with.
text-embedding-3-small model, which is priced at $0.02 per 1M tokens. A token is roughly three-quarters of a word, so a million tokens is a large pile of text — for a typical project, the first full read costs a few cents and the day-to-day top-ups round to nothing. The bill is yours and it is small.Step 3 — Run setup
Now back to Claude Code. The setup command gets the plugin's filing system ready for this project — it builds the shelves before anything goes on them. It won't ask you for your key yet; that comes in the next step. Run it and let it finish.
/knowledge-vault:setup
Step 4 — Run doctor, and paste your key
The doctor command checks that everything is wired up correctly, and it's the point where you hand over the OpenAI key from Step 2 — so have it on your clipboard before you run this. Doctor will tell you whether the key works and whether the plugin can reach OpenAI with it. If something's off, it says so here rather than failing silently later.
/knowledge-vault:doctor
Step 5 — Run index
This is the reading step — where the librarian actually goes through every shelf and files what it finds. The index command sends your conversations, notes, and memory to be read and catalogued. The first run on a project with a lot of history takes a minute or two because it's reading everything for the first time; that wait is the whole point, and it only happens once.
/knowledge-vault:index
Step 6 — Run search
The payoff. The search command takes a plain-English question and hands back the passage that answers it, along with where it came from. You don't need keywords or special syntax — ask the way you'd ask a colleague who'd read everything you'd written.
/knowledge-vault:search what did we decide about the pricing tiers
If that returns a relevant answer with a source attached, you're done — the plugin is installed and answering questions about your own work.
04/Using it day to day
From here on, only two commands matter. search whenever you need to remember something — a decision you can't fully reconstruct, or a number you know you wrote down somewhere three weeks ago. And index whenever you've added enough new material that you want it findable.
The reason you re-run index is worth understanding, because it's the one thing people trip over. The catalogue is a snapshot of what existed the last time it read. A conversation you had this morning isn't on the shelves until the librarian goes back through and files it. So if a search comes up empty on something recent, the usual cause isn't that the plugin missed it — it's that the plugin hasn't read it yet. Run index, then search again.
index at the end of a working session rather than the start of the next one. That way the catalogue is always current when you sit down, and you never lose a minute waiting on a read before you can ask your first question.05/Troubleshooting
Most problems come from one of four places, and all four have a quick fix. Find your symptom on the left.
| What you're seeing | Why it's happening | What to do |
|---|---|---|
| The install command can't find or access the plugin | You haven't accepted the GitHub invite, so the repository is still private to your account | Accept the invite from your email or GitHub notifications, then run the install command again |
| Doctor says it can't reach OpenAI, or the key won't work | The key is missing, was mistyped, or the OpenAI account has no billing set up | Run doctor again and paste the key carefully; confirm the key is active and billing is on in your OpenAI account |
| Search returns nothing, or nothing useful | You haven't run index yet, or you added material after the last read | Run index so the latest material is filed, then search again |
| Index seems to sit there for a while on a big project | The first run reads everything from scratch, which takes a little time | Let it finish — the full read happens once, and later runs only pick up what changed |
06/Your privacy
The plugin runs inside your own Claude Code, on your own machine, under your own OpenAI key. That last part is the whole privacy model, and it's worth being explicit about what it means.
When the plugin reads your material, the text is sent to OpenAI to be catalogued — using the key you created in Step 2, billed to your account, governed by your agreement with OpenAI. Tenkr is not in that loop: we never receive your key, we don't see your conversations or notes, and the questions you ask never reach us. There's no Tenkr server sitting between you and your own files, because there doesn't need to be.
Bring-your-own-key cuts the other way too: because the cost and the data both run through your account, you can see exactly what the plugin spends on your OpenAI dashboard, and you can revoke the key at any time to cut it off entirely. The plugin only has the access you choose to give it.