Skip to main content
On this pageOverviewThe jobWhat's insideIntegrationsInstall · kr 2 000FAQBetter together
Knowledge

Every session, note, and document — one searchable index.

Knowledge Vault indexes your Claude Code transcripts, your project memory, and a folder of your own documents into a local hybrid-search index, then answers natural-language questions with citations. Ask 'what did I decide about…' and get the answer plus where it came from.

  • Cited answers across your history
  • Index stays on your machine
  • Memory is read-only — never modified

kr 2 000one-time · lifetime updates · install →

/ 01 The job

What runs when you install the plugin.

Four commands, in order — provision, build, ask. The index is per-project and lives outside your repo; nothing is written into your project tree.

  1. 01setup wires Knowledge Vault into the current project — record-only, and safe to re-run.
  2. 02doctor checks and installs prerequisites — uv, Python ≥3.12, packages, and your OpenAI key — after one consent. Idempotent.
  3. 03index builds the local hybrid-search index over your transcripts, memory, and documents. Incremental by default, with a cost preview before large runs.
  4. 04search answers a natural-language question across the index and returns a synthesized, cited answer.
/ 02 What's inside

Every artefact you're getting.

A plugin is a folder. Here's everything in it — one skill, four commands, and the Python engine that does the indexing and search. Nothing hidden.

1 Skill

Atomic procedure Claude Code loads on cue, isolated context.

vault-searchFires when you ask about past sessions, memory, or your documents ("find where I…", "what did I decide about…").

The plugin sits at ~/.claude/plugins/knowledge-vault/, and the index it builds lives at ~/.claude/projects/<project>/knowledge-vault/ — both on your machine, nothing in your repo.

/ 03 Tools it reads + writes

Will it work with your stack?

Reads from
TranscriptsProject memoryDocuments folder
Writes to
Local search index
Requirements: Claude Code, plus uv and Python ≥3.12 (doctor provisions both). An OPENAI_API_KEY is required for embeddings — Keychain recommended on macOS. Optional OCR for PDFs/Office/images needs a MISTRAL_API_KEY via doctor --ocr.

Only declared calls leave your machine: chunk embeddings to OpenAI, and — if you enable OCR — documents to Mistral. The index, your memory, and your transcripts stay local. No Google Drive, no cloud storage.

/ 04 Install & price

Four steps. From buy to first run.

Buy on the right. On purchase, we invite your GitHub account to the private plugin repo you bought — the Tenkr marketplace itself is public — and email you the two commands below.

1Install Claude Code (Anthropic's free CLI). Already have it? Skip.
2Buy. We invite your GitHub account to the private plugin repo you paid for. The Tenkr marketplace itself is public.
3In Claude Code, run /plugin marketplace add TenkrAS/tenkr-plugin-marketplace to register the public catalogue. Exact command in your purchase email.
4Run /plugin install <plugin>@tenkr-plugin-marketplace — Claude Code clones it from your private repo and you're live. Type the plugin's command to start.
/ 05 Plugin-specific FAQ

What people ask before buying.

/ 01Does it send my documents anywhere?
Only chunk embeddings go to OpenAI (text-embedding-3-small); if you turn on OCR, documents go to Mistral. Nothing else leaves your machine — the index itself is a local SQLite file.
/ 02Does it modify my memory or notes?
No. Your project memory is read-only — indexed, never written.
/ 03Do I need an API key?
Yes — an OPENAI_API_KEY for embeddings. OCR is optional and needs a MISTRAL_API_KEY. doctor sets both up.
/ 04Does re-indexing charge me again?
No. Every chunk is cached by file hash, so unchanged content makes zero embedding calls on re-index.
/ 05Does it touch Google Drive or cloud storage?
No. It reads three local sources — your transcripts, your project memory, and a local documents folder. Nothing is pulled from Drive or any cloud service.