Skip to content
All docs
Docs

Geneziz Documentation

How to install, set up, and use Geneziz - a local-first knowledge base that captures your saves from X.com, GitHub, YouTube, Reddit and more into searchable Markdown, with an MCP server for AI assistants.

Overview

Geneziz is a local-first knowledge base that turns scattered saves into a structured, searchable second brain. It captures saves from X.com bookmarks, GitHub stars, YouTube, Reddit, Instagram, TikTok, Cosmos, Bluesky, Hacker News, and RSS feeds - plus any page you save from your browser - enriches each item with AI, and stores everything as plain Markdown files on your machine.

Each archived item gets a category, tags, a one-line description, and a priority rating, then lands in the desktop app (or web viewer) where you can browse, filter, and search it. A built-in MCP server exposes your knowledge base to AI assistants like Claude Code, Cursor, and Windsurf, so your coding sessions can query what you've saved.

Your data never leaves your machine. The files are grep-able, git-trackable, and readable in any text editor.

Installation

Download the latest desktop build for your platform. No Python or Node.js is required - the app ships as a self-contained native installer.

PlatformFileNotes
Windows x64.exe installer (from geneziz.app)Supported build
macOS.dmgNative universal build

After running the installer, launch Geneziz and the first-run setup wizard appears automatically.

First-run setup

The setup wizard collects four pieces of information. You can skip any of them and return later from Settings.

License key

Enter your license in GNZ-XXXX or GNZD-XXXX format. A license is $99 one-time and activates on up to 3 devices.

X.com

Install the Save to Geneziz browser extension (bundled with the Geneziz desktop app). It captures inside your own logged-in X.com session as you browse - Geneziz never reads your browser's stored cookies, and nothing is sent to any cloud. Sources that use official APIs (GitHub, YouTube) connect with your own token/OAuth.

AI provider

Choose how Geneziz enriches items. Four modes are supported:

  • Geneziz Cloud (proxy) - uses the Geneziz API, spends your included credits, no key needed
  • Anthropic - bring your own Anthropic key
  • OpenAI-compatible - OpenAI, OpenRouter, GLM, or any compatible endpoint
  • Ollama - fully local

Bringing your own key means unlimited processing with no credit consumption.

GitHub token

Optional. Paste a personal access token to sync your GitHub stars. Without it, the GitHub step is skipped.

Importing your data

Multiple accounts per source

Connect personal and work accounts side by side (GitHub tokens, YouTube OAuth, X.com session credentials) in Settings - Connections. Sync runs each active account; the viewer shows a combined view by default with per-account filtering.

Ask Geneziz (AI chat)

The built-in chat answers questions about your library in natural language, grounded in your indexed content. Conversations stay on your machine.

X.com bookmarks

X.com capture is bookmarks only, and it runs through the browser extension in your logged-in session (see First-run setup). CLI fetching loops over your connected sources and accounts:

bash
geneziz fetch --source github          # all active GitHub accounts
geneziz fetch --source github --account gh-work   # just one

GitHub stars

Sync your starred repositories. Each repo is categorized and filed as a Markdown knowledge file.

bash
geneziz sync

Pocket

Import a Pocket JSON export. Export your library from Pocket's web UI, then:

bash
geneziz import-pocket --input path/to/pocket-export.json

CLI reference

Geneziz ships with a full CLI. Run geneziz --help for the live list and flags.

CommandDescription
--versionShow version info
initCreate a config file
setupInteractive setup wizard
statusShow stats (bookmarks, tools, articles, media)
fetchCapture from your connected sources (multi-account aware)
processProcess pending bookmarks (AI or --no-ai)
syncSync bookmarks + GitHub stars
enrichAI-enrich minimal knowledge files
hydrateDownload media from tweets/articles
gapsDetect & repair knowledge-base gaps
indexRebuild the browse + search index
webLaunch the web viewer dev server
buildBuild the web viewer for production
serveServe the built viewer (no Node.js needed)
migrateMigrate an older library
dashboardLaunch the analytics dashboard
mcpStart the MCP server
search QUERYFull-text or --semantic search
ask QUESTIONQ&A over the knowledge base
insightsAI-generated insights
possibleAI idea-roadmap generator from your themes
skillInstall/show agent slash-command skills
vizTerminal dashboard for the knowledge base
runOne-command pipeline: fetch → process → index
resetDelete config and state, keep user data
exportExport the knowledge base to a zip backup
import-pocketImport bookmarks from a Pocket JSON export

The one-command pipeline runs the full cycle end to end:

bash
geneziz run             # fetch → process (AI) → index
geneziz run --no-ai     # deterministic filing only, no AI

AI providers

Geneziz supports four AI modes for enrichment:

ModeWhat it usesKey neededCredits
Proxy (default)Geneziz CloudNoneYes (~1 per item)
AnthropicAnthropicProvider keyNo
OpenAI-compatibleOpenAI / OpenRouter / GLM / etc.Provider keyNo
OllamaA local modelNoneNo

Bring your own key (BYOK) for unlimited, credit-free processing. With the default proxy, each AI-enriched item costs roughly one credit from your included balance.

Licensing & credits

Geneziz is $99 once - no subscription, ever.

  • License ($99 one-time): desktop app + CLI + MCP server + semantic search, 2,000 starter credits, up to 3 devices
  • Credit top-up packs: $12.99 (100) · $24.99 (250) · $44.99 (500) · $79.99 (1,000)
  • Skip credits entirely by bringing your own AI provider key

See /landing#pricing for current pricing.

Geneziz lets you search your knowledge base two ways - both run locally and offline.

  • Full-text search - fast keyword matching with ranking
  • Semantic search - finds items by meaning, not just exact words
bash
geneziz search "react performance"            # keyword
geneziz search "rust async runtime" --semantic # meaning-based

Your data

Everything is stored as Markdown with YAML frontmatter, directly on your machine. These are plain files you own - open them in any editor, search them with grep, or track them in git.

PathContents
bookmarks.mdChronological bookmarks, newest first
knowledge/tools/*.mdOne file per tool/repo
knowledge/articles/*.mdOne file per article
knowledge/media/Downloaded images and media

Sample tool file:

yaml
---
title: Tokio
type: tool
date_added: 2026-06-16
source: https://github.com/tokio-rs/tokio
tags:
  - rust
  - async
  - runtime
description: A runtime for writing reliable asynchronous applications with Rust.
---

The file body holds the README / notes for this item.

MCP server

The Model Context Protocol (MCP) server exposes your knowledge base to any MCP-aware AI assistant. It runs locally on your machine as a stdio server - there is no port to configure, and it reads your local files directly, so nothing leaves your machine.

Every tool requires an activated Geneziz account (the $99 one-time license). The server exposes 33 tools plus 4 knowledge:// resources (stats, index, trends, categories): keyword and semantic search, related-item and duplicate scans, recency and stats, listings (tools, articles, tags, categories, domains), single-item reads (tools, articles, bookmarks), ask_kb with a saved-conversation memory loop, and writes from your assistant (save_url, add_note, summarize).

The most-used tools:

ToolWhat it does
search_fulltextFast BM25 keyword search (exact names, errors, titles)
search_knowledgeHybrid keyword + semantic search (concepts)
get_tool / get_articleRead a full item by slug
find_relatedNearest neighbors of a file you trust
list_tools / list_articlesFiltered, paginated listings
get_recent / list_tags / list_domainsOrient inside the base
save_url / add_noteCapture links and write notes from your assistant

Full playbook - which tool to reach for, in what order - in Getting the Most Out of the Geneziz MCP Server. The complete tool inventory and per-client setup snippets are on the MCP page.

Setup

The one-command path: geneziz mcp register writes the server entry into every AI client it detects on your machine - Claude Desktop, Claude Code, Cursor, Windsurf, Codex, VS Code, ZCode, and OpenCode:

bash
geneziz mcp register

Each client gets a geneziz entry with the absolute path to your Geneziz executable, the mcp argument, and a GENEZIZ_DATA_DIR environment pin pointing at your data - so no PATH setup is needed. Registration only upserts Geneziz's own entry (other servers in the same file stay untouched, and a one-time backup is written before the first change), and it's easy to undo:

bash
geneziz mcp register --status        # what's registered, per client (JSON)
geneziz mcp register --remove --all  # remove only the Geneziz entries

The desktop app runs the same registration automatically on install and first run - default-on, with an opt-out in Settings.

Prefer to wire it by hand? Register the server with your AI assistant as a stdio MCP server using command geneziz with the mcp argument. For Claude Code, add a geneziz entry under mcpServers in ~/.claude.json:

json
{
  "mcpServers": {
    "geneziz": {
      "command": "geneziz",
      "args": ["mcp"]
    }
  }
}

Cursor and Windsurf accept the same server definition through their MCP settings UI. Copy-paste snippets for Claude Code, Cursor, Windsurf, and ZCode are on the MCP page.

In a new session, ask a question that touches your saves:

What tools have I bookmarked for Rust async?

The assistant searches your local knowledge base and answers with titles, categories, and snippets. See Setting up the Geneziz MCP server with Claude Code for the full walkthrough.

FAQ

Is my data really local? Yes. Everything lives as plain Markdown on your machine, works offline, is grep-able, and can be tracked in git.

Are there subscriptions? No. Geneziz is a single $99 one-time purchase.

Is it open source? Geneziz is proprietary software; the data format is 100% open - plain Markdown plus YAML.

Does it work on Mac and Windows? Yes. Geneziz ships as a native desktop app for macOS and Windows.

What happens if you shut down? Your files stay yours. There is no cloud dependency - every file remains readable in any editor, with or without Geneziz running.

Support

Questions, bugs, or feedback: support@geneziz.app. See /landing for product details and pricing.

On this page