Local-first isn’t nostalgia—it’s the fastest, most private way to work on a modern computer. When the core loop runs on your device, you skip logins, latency, and the “Are we connected?” tax. This starter kit gives you the practical baseline to make offline work—and on-device AI—feel instant.

1) Hardware: what actually matters

You don’t need a data center. You need balanced components and a clean setup.

Baseline (Laptop)

  • CPU/NPU: Any modern multi-core CPU. If available, choose an “AI PC”/Apple Silicon with an NPU/Neural Engine for efficient on-device inference.
  • Memory: 16 GB minimum. 32 GB if you’ll run local models or lots of browser tabs.
  • Storage: 1 TB NVMe SSD. Local beats cloud on throughput. Keep working sets on the internal drive.
  • Battery: Aim for 8+ hours real use. Offline should mean all-day.

Performance (Desktop/Workstation)

  • CPU: Recent 8–16 core.
  • GPU (optional): 8–16 GB VRAM if you plan to run 7–13B LLMs or heavy vision locally.
  • Memory: 32–64 GB.
  • Storage: Two NVMe drives (OS/apps, data/working sets) + large HDD/NAS for archives.
  • UPS: Small line-interactive UPS for graceful shutdowns and clean power.

Nice-to-haves

  • SD card/USB-C reader for frictionless ingest (photos, scans, docs).
  • LAN first: Wired Ethernet for big local transfers; Wi-Fi is fine for everything else.

2) OS setup: local by default, network by permission

File system & storage

  • One “Working” folder on your fastest NVMe (e.g., ~/Work). Keep active projects here so search and AI indexers stay hot.
  • Readable formats: Prefer markdown, CSV, PDF, plain images. Your future self will thank you.
  • Backups: 3–2–1 rule—3 copies, 2 media, 1 off-site. Local snapshot daily, external drive weekly.

Privacy & telemetry

  • Local account first: Avoid mandatory cloud logins for OS setup if the platform allows.
  • Telemetry: Turn off “diagnostics/experience improvement” where possible. Keep crash reports manual.
  • Firewall: Default deny for unknown apps. Allow only what you use.

Encryption & credentials

  • Full-disk encryption: Enable FileVault (macOS) / BitLocker (Windows Pro) / LUKS (Linux).
  • Password manager: Local-first (e.g., a vault file you keep and back up). Hardware key for admin logins.

Search & indexing

  • System search: Keep indexes on your NVMe “Working” folder for instant results.
  • Dev power: Pair ripgrep/fzf or an “Everything”-style indexer with keyboard shortcuts for lightning-fast find.

Network & sync

  • LAN > Cloud: For multi-device, start with peer-to-peer sync (e.g., local network). Only add internet sync if you truly need it.
  • Offline mode: Get comfortable toggling Wi-Fi off. Great for focus, battery, and privacy.

3) Apps that shine offline

  • Mail & Calendar: Use clients that keep full local copies (and can be searched offline). CalDAV/CardDAV or direct IMAP are your friends.
  • Notes & docs: Markdown/HTML + local images. Version with simple folders or git if you’re technical.
  • Reference library: Save key docs/pages as PDFs. Tag and index locally.
  • Media & scans: Use your phone as a local scanner (save to your LAN/desktop instantly).

4) On-device AI: practical settings

Local AI should feel like power steering, not a spectacle.

  • Model size: Laptops: 3–8B models for chat/summarize/transcribe. Desktops with a mid-range GPU: 7–13B with good speed.
  • Quantization: Start with a quality-focused setting (e.g., Q6) and drop to a lighter one (e.g., Q4) if you need more speed.
  • Context length: Use what you’ll actually need (8k is fine for most tasks). Longer contexts cost RAM/VRAM.
  • Privacy: Keep model caches and embeddings on your encrypted drive. Don’t index folders you wouldn’t email to yourself.
  • Web search: Leave off by default. When you enable it, log what was fetched and keep citations.

5) Daily practices that make offline sing

  • Morning sync, then airplane mode: Pull the latest mail/files first thing. Work offline until lunch. Re-sync mid-day if needed.
  • Single source of truth: Decide where each data type “lives” (mail, tasks, notes, docs). Avoid duplicating the same item across five apps.
  • Save locally, then share: Draft everything on disk, push outward when done.
  • Weekly export ritual: One time box to export critical mail threads, calendars, and notes to your archive folder.
  • Battery hygiene: Calibrate once a month. Keep spare chargers where you work.

6) Troubleshooting playbook

  • It’s slow: Check free space (aim for 20%+). Move the project onto NVMe. Pause background indexers.
  • Search misses files: Rebuild the index on your “Working” folder only.
  • App nags for cloud: Look for “Store locally” or an export path. If missing, replace the app.
  • Model runs hot: Lower context, drop quantization, or switch to a smaller model for daily tasks.

7) The 10-minute Local Setup (checklist)

  • ✅ Create ~/Work on NVMe; move active projects there
  • ✅ Enable full-disk encryption
  • ✅ Turn off telemetry/“experience improvement” toggles
  • ✅ Configure offline mail & calendar with full local copies
  • ✅ Install a fast local search/indexer
  • ✅ Set up daily snapshots + weekly external backup
  • ✅ Install an on-device model for summarize/transcribe
  • ✅ Add a keyboard shortcut to toggle Wi-Fi
  • ✅ Pick one P2P/LAN sync for multi-device (optional)
  • ✅ Write a 1-page “Where my data lives” note

Local-first isn’t dogma—it’s the sensible default

Use the cloud when it clearly helps (team sync, shared search, heavy retraining). For everything else, keep the loop on your machine: faster, quieter, more private, and less brittle. That’s how we build at Clairos—sovereign by design, local by default.