The Last Light Survival Guide

A complete, fully-offline survival library — 27 sections, 33 reference books, an AI that cites them, and offline maps. Runs off a USB stick. Free.

See it in action

Recorded in a real browser against a local model: open the guide, ask Bunker Bot, and it answers from the bundled books with inline source citations — fully offline.

A comprehensive, dependency-free survival reference built for the grid-down scenario. Zero internet required after download. Works from a USB drive, laptop, tablet, phone, or a Raspberry Pi. No accounts, no tracking, no ads.

• 🌐 Live: https://bunkerbot.co.uk · Download (free): https://bunkerbot.co.uk/download.html
• 🤖 Offline AI that runs locally (Ollama / llamafile) and cites its sources — it won't invent drug doses.
• 📚 33 bundled reference books + browser-side full-text search, all offline.
• 🗺️ Offline maps (PMTiles) and one-click Kiwix expansion for all of Wikipedia & Project Gutenberg.
• 🔧 Pure static HTML/CSS/JS — no build step, no server, runs from file://.

⭐ If this is useful, star the repo — it helps others find it. Mirrors and re-shares encouraged (CC BY-SA).

🩺 Help wanted — expert review

The content is broad and AI-assisted, and the medical material has not yet been reviewed by a licensed clinician. The inline citations exist so you can verify every claim. If you're a doctor, paramedic, nurse, vet, midwife, ham, or experienced homesteader, please tear apart the relevant section — open a medical/content correction issue. That review is the single most valuable contribution right now.

---

Quick Start

1. Copy this folder to a USB drive or local directory
2. Open index.html in any modern web browser
3. Bookmark it — Ctrl+D / Cmd+D

Do this now: Print quick-reference.html and cards.html before an emergency. Laminate the wallet cards.

Make a verified USB stick (one command)

sh make-usb.sh /Volumes/MYUSB           # macOS
sh make-usb.sh /media/$USER/MYUSB       # Linux
sh make-usb.sh /Volumes/MYUSB --lite    # app only, ~4 MB (no PDFs/maps/AI index)

It copies the guide, then re-reads every byte on the stick against fresh SHA-256 hashes — so a flaky drive or bad copy is caught now, not the day you need it.

Raspberry Pi — one-line survival server

Turn any Pi (or Debian-ish box) into a self-contained survival AI appliance — guide + books + maps + a local model, served to your whole network on boot:

curl -fsSL https://raw.githubusercontent.com/hamtechautomation/survivalai/main/pi-setup.sh | sh

Then open http://<pi-address>:8080 from any phone or laptop on the network. It downloads the full guide, installs Ollama, pulls a model sized to the Pi's RAM, and installs a systemd service. Re-running is safe; each step skips what's already done.

---

Setting Up Bunker Bot (Offline AI Assistant)

Bunker Bot connects to Ollama running on your local machine. It is completely optional — the guide is fully usable without it. When Ollama is running, click the Bunker Bot button on any page to open the chat panel.

Bunker Bot features:

• Grounded in the bundled library — answers are retrieved from the 33 bundled PDFs (BM25 search over ~15,000 text chunks) and cited inline with source + page, shown as chips under each answer. Works on every page.
• Local (Ollama) or cloud (Claude API) — toggle in the panel
• Model selector, temperature slider, token counter
• Emergency mode (direct, step-by-step answers with no hedging)
• Copy last response / Export full chat; streaming responses with a stop button

macOS / Linux

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a model (phi3:mini fits in 4 GB RAM)
ollama pull phi3:mini

# Start with CORS open (required when opening from file://)
OLLAMA_ORIGINS=* ollama serve

To run Ollama automatically on login (macOS), add to ~/.zshrc:

export OLLAMA_ORIGINS='*'

Then reload: source ~/.zshrc

For Linux systemd, edit the service:

sudo systemctl edit ollama.service --force

Add:

[Service]
Environment="OLLAMA_ORIGINS=*"

sudo systemctl daemon-reload && sudo systemctl restart ollama

Windows 10 / 11

1. Download the Ollama installer from ollama.ai
2. Run the installer (adds Ollama to system tray)
3. Open PowerShell:

$env:OLLAMA_ORIGINS = "*"
ollama pull phi3:mini
ollama serve

To set OLLAMA_ORIGINS permanently: System Properties → Environment Variables → New User Variable → OLLAMA_ORIGINS = *

USB / Air-Gapped Machine

1. On a connected machine: ollama pull llama3:8b (downloads to ~/.ollama/models/)
2. Copy the entire ~/.ollama folder to USB
3. On the air-gapped machine: install Ollama, copy the ~/.ollama folder back
4. Run OLLAMA_ORIGINS=* ollama serve — models load from the copied folder

Recommended Models by Hardware

Model	RAM needed	Speed	Quality	Notes
phi3:mini	4 GB	Fast	Good	Best for constrained hardware
gemma3:4b	6 GB	Fast	Good	Excellent instruction following
llama3:8b	8 GB	Medium	Very good	Best balance of speed/quality
mistral:7b	8 GB	Medium	Very good	Strong at structured output
llama3:70b	48 GB+	Slow	Excellent	Needs a workstation

---

Terminal Mode — Bunker Bot with No Browser

Prefer the command line, or on a headless box (a Pi, a server, SSH)? bunkerbot.py is the whole assistant in one zero-dependency Python file — the same BM25 retrieval over the bundled books, the same source citations, no browser required.

python3 bunkerbot.py                       # interactive guide
python3 bunkerbot.py "how do I stop heavy bleeding"   # one-shot answer
chmod +x bunkerbot.py && ./bunkerbot.py    # then just ./bunkerbot.py

It auto-detects a provider: Ollama (localhost:11434), a llamafile (localhost:8080), or Claude (set ANTHROPIC_API_KEY). With no model running it drops to search-only mode — still a fast, citing lookup tool over the library.

• /emergency — terse, critical-action-first mode · /help · /quit
• Env: BUNKERBOT_MODEL, BUNKERBOT_TEMP, NO_COLOR

Python 3 standard library only — nothing to install. Runs on a Raspberry Pi.

Content Overview

28 Content Sections

Section	Key Topics
Food & Water	Purification, preservation, foraging, passive fishing, sanitation
Medical & First Aid	Trauma, wound care, meds, rabies, tetanus, anaphylaxis, eye injuries
Energy	Solar sizing, battery banks, hand tools, fuel storage
Shelter & Construction	Expedient shelter, insulation, earthworks, roofing
Communications	Ham radio, Morse code, signal mirrors, frequency reference
Navigation & Maps	Celestial navigation, star charts, map reading, dead reckoning
Security & Defense	Physical hardening, bug-out decision tree, threat assessment
Knowledge & Literacy	Education continuity, skill transfer, library curation
First Principles	Periodic table, physics/chemistry, key equations, the invention method
Agriculture	Seed saving, planting calendars, soil health, crop rotation
Animal Husbandry	Chickens, goats, rabbits, bees, housing and feeding
Veterinary Care	Vital signs, diseases by species, parasites, birthing, biosecurity
Pregnancy & Infant Care	Prenatal, labour, birth danger signs, newborn & infant care
NBC / EMP Threats	Radiation exposure limits, decontamination, Faraday cages
Disaster Playbooks	Hurricane, earthquake, wildfire, pandemic, nuclear event
Climate & Regional	Zone-by-zone survival, UK foraging calendar, wild edibles
Metallurgy	JABOD forge, blacksmithing, smelting, nail making
Governance	Community organization, conflict resolution, resource allocation
Psychology & Morale	Stress management, group dynamics, box breathing, grief
Chemistry & Materials	Soap making, bleach production, fermentation chemistry
Textiles & Clothing	Spinning, weaving, leather tanning, pattern cutting
Vehicles & Transport	Fuel stabilization, EMP hardening, bicycle maintenance
Salvage, Rebuild & Fuel	Parts salvage & safe stripping, generator maintenance/rebuild, battery salvage (lead-acid, 18650), biodiesel, SVO, ethanol, wood gas
Power Generation	Pedal generators, water wheels, wind, biogas, steam
Building & Structures	Cob, earthbag, timber frame, defences, site selection
Medicine Making	Tinctures, ORS, antiseptics, medicinal plants
Water Systems	Rainwater harvesting, wells, gravity distribution, irrigation
Practical Projects	Cooling/aircon, air filters, dams & hydro, pumps, toilets, radiation checking, mills, breeding, food storage
Practical Skills	36 step-through visual skill guides across 7 categories

Practical Skills Section

skills.html provides interactive step-by-step guides for hands-on survival skills. Each skill includes:

• Guided step-through mode with keyboard navigation (← →)
• Inline SVG diagrams for key techniques
• Built-in countdown timers for timed steps
• CPR metronome (110 bpm, auto-pauses at 30 for breaths)
• Practice session logging and competency rating (None → Getting There → Competent → Can Teach)
• Instructor notes with common mistakes and fast-track tips
• Printable step cards

Skill categories: Fire (9) · Knots (11) · Navigation (5) · First Aid (4) · Hunting & Fishing (4) · Plant ID (3) · Construction (4, incl. blade sharpening)

Progress is stored locally via localStorage and can be exported/imported as JSON for backup or sharing across devices.

14 Interactive Tools (all offline, no server required)

Tool	Function
Calorie Calculator	BMR + activity-level daily calorie target
Water Budget	Per-person daily water by climate and activity
Solar Sizing	Panel wattage, battery amp-hours, days of autonomy
Supply Tracker	Days-remaining calculator with red/amber/green status
Skill Inventory	Track and score your group's competencies
Planting Calendar	USDA zone-aware planting and harvest dates
Bug-Out Decision Tree	Structured stay/go decision with route assessment
Radio Frequency Reference	Pre-loaded emergency frequencies by type
Medication Reference	Dosing and storage guide for common emergency meds
Antibiotic Dosing Calculator	Weight-based pediatric/adult dosing for 8 antibiotics, tablets/mL per dose
Survival Timer & Priorities	Rule of 3s + countdown timers (boil/bleach/CPR/tourniquet) with beep
Celestial Latitude Finder	Latitude from Polaris or the noon sun — no GPS
Firewood & Heating Estimator	Wood needed to heat a space (lb/day, cords)
Crop Spacing & Seed Planner	Plants per bed and seeds to sow, with germination margin

Other Pages

• quick-reference.html — 2-page printable emergency card covering the first 72 hours
• cards.html — 16 wallet-sized lamination cards (water purification, Morse, wound care, dental, solar troubleshooting, etc.)
• gear.html — Interactive gear checklist with localStorage persistence
• ai-setup.html — Step-by-step Bunker Bot / Ollama setup with troubleshooting
• expansion.html — Expansion Library: one-click links to download all of Wikipedia, Project Gutenberg, and offline maps (Kiwix/ZIM + OSM)
• maps.html — Interactive offline browser map (PMTiles) with a region picker; works on a minimal static server, no internet
• index-print.html — Printable master index of the whole archive (for paper)

---

File Structure

last-light-survival-guide/
├── index.html                  ← Dashboard — start here
├── about.html                  ← About the project (ethos, credits, get involved)
├── download.html               ← Download packages + install guides
├── skills.html                 ← Practical Skills — 36 interactive guides
├── tools.html                  ← 14 interactive calculators
├── quick-reference.html        ← 72-hour emergency card (print first)
├── cards.html                  ← 16 wallet cards (laminate these)
├── gear.html                   ← Gear checklist (localStorage)
├── literature.html             ← Reference library
├── expansion.html              ← Expansion Library (Wikipedia/Gutenberg/maps)
├── maps.html                   ← Offline browser map (PMTiles) + region picker
├── index-print.html            ← Printable master index
├── ai-setup.html               ← Bunker Bot / Ollama setup guide
├── changelog.html              ← Version history
├── manifest.json               ← PWA manifest
├── sw.js                       ← Service worker (offline caching)
├── offline.html                ← Offline fallback page
├── verify.sh                   ← Check a copied archive is intact (SHA-256)
├── make-usb.sh                 ← One-command verified USB copy
├── pi-setup.sh                 ← Raspberry Pi one-line server bootstrap
├── bunkerbot.py                ← Terminal Bunker Bot (zero-dependency CLI)
├── MANIFEST.sha256             ← Checksums for verify.sh
├── make-torrent.sh             ← Build a torrent for P2P distribution
├── get-knowledge.sh            ← Download a Kiwix ZIM (Wikipedia/Gutenberg…)
├── extract-map.sh              ← Carve any region of the world into a map file
├── deploy-s3.sh                ← Publish to AWS S3 + CloudFront
├── build-packages.sh           ← Build the downloadable ZIPs
├── rebuild-index.sh            ← Rebuild the PDF search index after adding books
├── sections/                   ← 28 content section pages
│   ├── food.html               medical.html        energy.html
│   ├── shelter.html            communications.html navigation.html
│   ├── security.html           knowledge.html      science.html (First Principles)
│   ├── agriculture.html        animal.html         veterinary.html
│   ├── maternal.html           nbc.html            disasters.html
│   ├── climate.html            metallurgy.html     governance.html
│   ├── psychology.html         chemistry.html      textiles.html
│   ├── vehicles.html           salvage.html           build-power.html    build-structures.html
│   ├── medicine-making.html    build-water.html    projects.html (Practical Projects)
├── pdfs/                       ← 33 bundled reference PDFs (gitignored; on the USB)
│   ├── index.html              ← PDF library browser
│   └── extract.py              ← Rebuilds the PDF search index
├── maps/                       ← Offline map data (.pmtiles, gitignored; on the USB)
├── search/
│   ├── search-index.json       ← Client-side page search index
│   └── pdf-chunks.json         ← Full-text PDF search index for Bunker Bot (~29 MB)
└── assets/
    ├── css/style.css           ← All styles (dark, hi-contrast, night vision)
    ├── icons/icon.svg          ← PWA app icon
    ├── vendor/maps/            ← Leaflet + protomaps-leaflet + pmtiles (offline map renderer)
    └── js/
        ├── app.js              ← Navigation, sidebar, SVG icons, PWA setup
        ├── search.js           ← Client-side full-text search
        ├── bunker-bot.js       ← Bunker Bot panel, Ollama/Claude, RAG + citations
        ├── librarian.js        ← PDF library + BM25 retrieval for Bunker Bot
        ├── tools.js            ← Calculator functions
        ├── shared-progress.js  ← localStorage progress tracking
        ├── skills-data.js      ← Skill definitions and SVG diagrams
        └── skills.js           ← Practical Skills UI engine

---

Technical Notes

Offline / file:// Protocol

The guide is designed to work from file:// — no server required. Two browser restrictions apply:

Feature	file://	http://localhost
localStorage	✅ All browsers	✅ All
Service worker (PWA)	❌ Not supported	✅ All
Bunker Bot (Ollama)	⚠️ Needs CORS flag	✅ All

To serve locally (enables PWA caching):

python3 -m http.server 8080 --directory /path/to/last-light
# Open: http://localhost:8080

Bunker Bot CORS fix for file://:

OLLAMA_ORIGINS=* ollama serve

Display Modes

Mode	When to use
Normal (dark)	Default — comfortable in low light
Hi-C	Maximum contrast — visually impaired users
NV (night vision)	Red tones — preserves scotopic vision outdoors

Toggled via the pill bar fixed to the bottom of every page. Persisted across sessions.

Font size (Normal / A+ / A++) is also persisted.

Keyboard Shortcuts

Key	Action
/	Focus search
Escape	Close sidebar / Bunker Bot panel / search
Ctrl+P	Print current page
← →	Navigate steps in Practical Skills modal
Tab	Skip-nav link (first Tab press on any page)

Print

Every page has a full @media print stylesheet. Sidebar, Bunker Bot panel, and UI chrome are hidden. Page numbers and the guide title print in the footer.

Special print pages:

• quick-reference.html — formatted for 2-page US letter
• cards.html — formatted for wallet card stock
• Practical Skills — each skill can be printed as step-by-step cards via the Print button in the skill modal

PWA / Install

On supported browsers (Chrome, Edge, Safari iOS), the guide can be installed as a Progressive Web App. An install prompt appears on second visit on mobile. Once installed, it works fully offline with no browser chrome.

---

Expanding the Library — Wikipedia, Gutenberg & Maps

The guide app itself is tiny (~2 MB); with the ~30 MB AI index it still fits anywhere. When you have internet, the Expansion Library (expansion.html) turns it into a launchpad for the rest of recorded knowledge — all read offline afterwards:

• Wikipedia & Project Gutenberg via Kiwix ZIM files — medical Wikipedia (~1–2 GB), Simple Wikipedia, full Wikipedia (maxi ~100 GB / nopic ~55 GB / mini ~10 GB), all 70k+ Gutenberg books, iFixit, Wiktionary and more. get-knowledge.sh <zim-url> downloads one (resumable).
• Offline maps. maps.html renders OpenStreetMap data from a single .pmtiles file directly in the browser — no map server, no per-view tile downloads, no internet. It uses Leaflet + protomaps-leaflet (Canvas 2D, runs on a Raspberry Pi) and loads the whole regional file into memory with one plain GET, so it needs no HTTP byte-serving and works on python -m http.server. A region picker switches between downloaded maps.

Download a map of anywhere:

# carve any bounding box out of the global map (megabytes, not the 100 GB planet)
sh extract-map.sh <name> <minLon> <minLat> <maxLon> <maxLat> [maxzoom]
# example — Greater London at street detail:
sh extract-map.sh london -0.55 51.25 0.30 51.70 14

The script auto-downloads the pmtiles tool and pulls only your box via HTTP range requests, then drops maps/<name>.pmtiles into the folder where the map page's region picker will find it. (maps/*.pmtiles and pdfs/*.pdf are gitignored — large data travels on the USB copy, not in git.)

Maps must be served over HTTP (even python3 -m http.server works) — browsers block reading large local files from a raw file:// path.

---

Verifying Your Copy

After copying the folder to a USB drive or another machine, confirm nothing was corrupted or lost in transit:

sh verify.sh

This re-computes a SHA-256 for every catalogued file and checks it against MANIFEST.sha256. ✓ All N files present and unchanged. means the copy is intact; anything else lists the files that are missing or altered.

To regenerate the manifest after intentionally changing files:

find . -type f ! -path './.git/*' ! -path './.claude/*' ! -path './node_modules/*' \
  ! -name '.DS_Store' ! -name 'MANIFEST.sha256' \
  | sed 's|^\./||' | LC_ALL=C sort \
  | while IFS= read -r f; do shasum -a 256 "$f"; done > MANIFEST.sha256

---

Sharing It Widely (Censorship-Resistant Distribution)

This guide is meant to spread freely and survive takedowns. Two peer-to-peer methods let it propagate with no central server to block:

BitTorrent

sh make-torrent.sh

This builds a .torrent of the whole folder (needs mktorrent or Transmission's transmission-create). Open the resulting file in any BitTorrent client and seed it; then share the .torrent file or its magnet link anywhere. As long as one person seeds, the entire archive — including the PDF library — stays downloadable by anyone. Nothing is published until you choose to seed.

IPFS

If you run IPFS:

ipfs add -r --quieter .

The final hash it prints is a permanent content address. Anyone can fetch the whole guide with ipfs get <hash>, or view it through a public gateway at https://ipfs.io/ipfs/<hash>. The content is addressed by its hash, so it can't be silently altered, and it stays reachable as long as any node pins it.

Both methods pair well with the MANIFEST.sha256 check above: recipients can verify what they received is byte-for-byte the original.

---

Hosting It Publicly (AWS S3 + CloudFront)

S3 is a good fit because, unlike GitHub Pages, it has no per-file size limit — it holds the large PDFs and map files too. The guide is plain static files, so hosting is just an upload.

Important: serve it over HTTPS. The offline features (service worker, PWA "install", offline-after-first-visit) and the map's "Locate me" GPS only work over HTTPS or localhost. A raw S3 website endpoint is HTTP-only — so put CloudFront in front of the bucket. Also host at the domain root, not a sub-path (the service worker pre-cache list uses root-absolute paths).

One-time setup

1. Create an S3 bucket (e.g. my-survival-guide). Keep it private (block public access on) — CloudFront will read it via Origin Access Control.
2. Create a CloudFront distribution with that bucket as the origin:
   • Origin access: Origin Access Control (OAC), and let it update the bucket policy so only CloudFront can read.
   • Default root object: index.html
   • Viewer protocol policy: Redirect HTTP → HTTPS
   • (Optional) attach your domain + an ACM certificate (in us-east-1).
   • (Optional) Custom error response: 403/404 → /offline.html (200).
3. Configure the AWS CLI locally: aws configure with write access to the bucket.

Deploy / update

BUCKET=my-survival-guide  CF_DIST=E1234ABCD  sh deploy-s3.sh

deploy-s3.sh syncs the whole folder (including the gitignored pdfs/ and maps/, since they're on your disk), sets long cache for big assets, no-cache for sw.js so updates propagate, and invalidates CloudFront if CF_DIST is set.

Caching & how updates reach people

Everything is cached for 1 year (cheap, fast — deploy-s3.sh sets it). That's safe even though filenames aren't content-hashed, because the service worker is the update channel:

• sw.js is served no-cache, so browsers always re-check it.
• When you change site content, bump the cache version in sw.js (e.g. last-light-v30 → v31). The new worker installs with cache:'reload', which force-refreshes every file past the year-long browser cache, then takes over. Returning visitors update on their next load.
• deploy-s3.sh also invalidates CloudFront so the edge serves fresh copies immediately. The one rule: bump the version when you deploy a content change.

Good to know

• Cost: storage is trivial (~$0.02/GB·month). The spend is CloudFront egress — the ~29 MB AI index loads when someone uses Bunker Bot, and a ~60 MB map loads when someone opens maps.html. Fine for modest traffic; watch it if you go viral.
• Bunker Bot's local-AI mode won't work for remote visitors (it calls their localhost Ollama) — by design; it degrades gracefully and offers Claude mode. The cited PDF retrieval still runs entirely in the visitor's browser.
• Maps work either way — the viewer loads the whole .pmtiles with one GET, and S3 also supports range requests if you later switch to streaming.

---

Contributing

No build system, no dependencies. To add a section:

1. Copy an existing file from sections/
2. Add an entry to NAV_PAGES in assets/js/app.js
3. Add the page path to PRECACHE in sw.js
4. Add search entries to search/search-index.json

All icons are inline SVG strings defined in assets/js/app.js (window.ICONS). No icon fonts, no CDN calls, no external resources of any kind.

---

License

Dual-licensed — see LICENSE and LICENSE-CODE:

• Code (HTML/CSS/JS, scripts, infra templates): MIT (LICENSE-CODE)
• Content (the written survival/reference material): Creative Commons Attribution-ShareAlike 4.0 (CC BY-SA 4.0)
• Bundled books/data: retain their own licenses (public domain or CC); the Hesperian guides are CC BY-SA non-commercial — keep them free.

You are free to share and adapt the content for any purpose, even commercially, as long as you give attribution and distribute your contributions under the same license.

Public domain sources incorporated:

• FEMA wood gasifier plans
• US Army Field Manual 21-76 (Survival)
• Hesperian Health Guides (CC BY-SA)
• Nuclear War Survival Skills — Cresson Kearny / OISM (public domain)
• USDA planting zone data
• FCC emergency frequency allocations

---

The Last Light Survival Guide — v1.0.0 — June 2026
Lite app ~2 MB · full bundle (33 PDFs + AI index + a regional map) ~580 MB · fits any pen drive — and expands to all of Wikipedia, Gutenberg and world maps on demand.