knowledge-vault

Turn any Obsidian vault (or plain folder of markdown) into a first-class memory for Claude Code — search it, save your work to it, keep it versioned and indexed, and let every agent share it. Config-driven and universal: point it at your vault and go.

---

Install

The knowledge-base plugin is pure Claude Code content — no CLI, no build step, no dependencies. Two commands to install, one to configure. Run the two install commands one at a time (paste the first, hit enter, wait, then the second):

/plugin marketplace add Elnora-AI/knowledge-vault

/plugin install knowledge-base@knowledge-vault

Then point it at your vault:

/kb-setup

/kb-setup asks for your vault path, writes the (gitignored) config, creates the default folders, and verifies everything. That's it — the full experience is live.

Prefer to configure by hand? Copy settings.example.md to .claude/knowledge-base.local.md in your project and set vault_path. See that file for every option.

Using Codex, Cursor, or another agent

The slash commands and dispatched subagents are Claude-Code-only, but the conventions are portable. Drop AGENTS.md at your project root and any agent can read/search/save to your vault following the same rules.

---

What you get

• Search your vault — the vault-access skill teaches the agent to find documents fast with Glob/Grep/Read across .md, .base, and .canvas files, using your frontmatter and folder conventions.
• Save your work — /md turns pasted text into a clean, frontmattered, cross-linked vault document; /note fetches a URL, summarizes it, and files it; /update-knowledgebase persists a whole session into the right folders via the vault-curator agent.
• A task system — /task, /tasklist, /task-done, /task-triage implement a five-file task model (inbox → to-do → in-progress → done / cancelled) with Obsidian Tasks syntax, due dates, and priorities.
• Obsidian power tools — the knowledge-agent writes Flavored Markdown (callouts, embeds, block refs, Mermaid), builds Bases (.base) database dashboards, and draws Canvas (.canvas) diagrams. Four reference docs ship the full syntax.
• Auto-maintained index — a session-start hook rebuilds a master index of every document, grouped by folder, so the agent always knows what already exists. A write hook keeps it fresh.
• Document versioning — new versions instead of overwrites, with supersedes/superseded_by frontmatter links.
• Cross-agent coordination — the agent-coordination skill lets any agent pull vault context on demand (the "files as context" pattern).

Everything is config-driven: folder names, task paths, the index location, and the default owner all come from your config. Nothing is hardcoded, so your vault stays yours.

Slash commands

Command	Does
/kb-setup	First-run: write config, create folders, verify
/md [text]	Format pasted text into a vault document with frontmatter + cross-links
/note [url]	Fetch a URL, summarize it, save as a dated note
/task <desc>	Add a task (due date, priority, section)
/tasklist [filter]	Query tasks (today / overdue / this week / priority / search)
/task-done <desc>	Move a task between states
/task-triage	Review the inbox one task at a time
/update-knowledgebase	Persist this session's results into the vault

Skills & agents

• Skills: vault-access (search/create), file-naming (naming conventions), agent-coordination (cross-agent access).
• Agents: knowledge-agent (fetch/create/version, .base, .canvas), vault-curator (files a session's results into the right folders).

---

Configuration

All paths live in .claude/knowledge-base.local.md (gitignored — each machine has its own). The only required field is vault_path. Sensible defaults cover everything else:

---
vault_path: /path/to/your/obsidian/vault   # REQUIRED — absolute path
vault_dir: ""                              # optional subfolder that is the vault root
task_inbox: tasks/inbox.md                 # the five task files
task_todo: tasks/to-do.md
task_in_progress: tasks/in-progress.md
task_done: tasks/done.md
task_cancelled: tasks/cancelled.md
notes_dir: notes                           # catch-all notes; holds the auto-index
index_file: notes/index.md
vault_title: Knowledge Base
default_owner: ""
---

Default folder layout (rename or extend freely — it's all config):

notes/  tasks/  meetings/  people/  policies/  projects/  reference/

Full field reference: settings.example.md.

---

Connectors (optional)

Want meeting transcripts, call notes, or any dated records to flow into your vault automatically? The connectors/ directory ships a small, source-agnostic sync framework — a Source adapter plus a SyncEngine that formats each record and writes it into the vault — with a Quill reference adapter as a worked example. Bring your own tool (Granola, Otter, Quill, …) by implementing one small Source interface.

On top of the core pipeline, connectors optionally give you the full automation loop:

• LLM formatting — a two-call pipeline (metadata + complete verbatim transcript) that classifies each record against your routes, tags it, and extracts action items. Degrades to passthrough without an API key.
• CRM linking — keep a CSV CRM inside your vault in sync with your meetings: participants are matched by email, last-contact fields stamped, unknown externals auto-created (organizations too), and per-call notes appended. Column-driven — any schema works. Pairs perfectly with the CRM that google-workspace scaffolds.
• Action items → task inbox — extracted follow-ups land in your task inbox with due dates resolved and the transcript linked.
• Verification — an auditable verify command that catches missing or truncated transcripts.
• Scheduling — install-schedule registers the sync on launchd (macOS), Task Scheduler (Windows), or cron (Linux). Set-and-forget.

Using Quill? The full automation needs only two installs — the Quill desktop app and this plugin. The connector reads Quill's local database directly (no API keys, no MCP server needed for syncing); one config file plus install-schedule and every meeting flows into your vault on a schedule. Quill's own MCP bridge can additionally be registered in .mcp.json for interactive meeting search — optional and independent of the sync. Step-by-step: connectors/README.md → Quick start (Quill).

See connectors/README.md. This is entirely optional and off by default; the core plugin never depends on it.

---

Supercharge with google-workspace

Pair this with Elnora-AI/elnora-google-workspace and the two become one self-driving system: Gmail and Calendar feed a lightweight CRM that lives right inside your vault, kept fresh on a schedule with no manual upkeep. The vault is the memory; google-workspace is the senses.

Once both are installed:

gw auth setup                 # one-time Google sign-in
gw crm init                   # scaffold contacts.csv in <vault>/crm
gw gmail sync-crm-install     # schedule email → CRM (mac/Windows/Linux)
gw calendar sync-crm-install  # schedule calendar → CRM

gw crm init reads the vault_path from your .claude/knowledge-base.local.md — no extra configuration. Both plugins still work entirely on their own; this just makes them better together.

---

Safety

Read SAFETY.md. In short: a write hook only ever reads your config, the plugin never deletes or overwrites documents (it versions them), your real vault_path is gitignored, and no secrets are stored or transmitted. The plugin's scripts and hooks talk to your local filesystem only; the sole exception is /note, which uses Claude Code's built-in WebFetch on the URL you provide.

---

Part of the Elnora family

Open-source agent tooling from Elnora AI — free, universal, config-driven tools that wire Claude Code (or any AI coding agent) into the systems you run your company on. Each works 100% standalone; install several and they chain into end-to-end workflows.

• elnora-linear — Linear issue management — search, bulk edit, agents, and a config-driven curator
• elnora-slack — the entire Slack Web API as a CLI plus agent skills with a draft-and-approve send gate
• elnora-whatsapp — read, search, and send WhatsApp from your own paired account, 100% local
• elnora-google-workspace — Gmail, Calendar, Drive, Docs, Sheets, Forms, Tasks, plus any Google API via Discovery
• elnora-merit-aktiva — Merit Aktiva accounting and Merit Palk payroll as a CLI and plugin
• elnora-vanta — read-only Vanta compliance — frameworks, tests, controls, and vulnerabilities as agent-friendly JSON
• elnora-luma — Luma (lu.ma) events — all 61 public API endpoints as a spec-driven CLI with safety guardrails
• elnora-travel — a real travel agent — live flights, hotels, Airbnb, Booking.com, and routes in one itinerary
• elnora-websearch-tools — web search — Exa, Tavily, Perplexity, Firecrawl, and Valyu CLIs and skills in one plugin

License

Apache-2.0. Maintained by Elnora AI. Contributions welcome — see .github/CONTRIBUTING.md.