FallHub · get started

Deploy FallHub · 20 minutes.

Fork the repo. Pick a vertical template. Fill the kernel. Bring your own AI key. Slide the autonomy dial from Watch to Auto as trust grows. Zero servers to rent, zero data to give up, zero monthly rent unless you want optional support.

Step 1 · Fork

Fork the FallHub repo.

Click Fork on GitHub. Name it after your business. GitHub Pages auto-deploys within a couple of minutes.

You now have your own live FallHub at https://YOUR-USERNAME.github.io/YOUR-REPO-NAME/.

Zero-setup version

Prefer to skip cloning entirely? Use the hosted reference. All state lives in your browser (IndexedDB) — nothing on our servers. You lose the ability to customise the code but gain zero-friction start.

Step 2 · Pick vertical

Choose the template.

Open ai/kernel.json in your forked repo. Replace the example content with a template from verticals/:

Step 3 · Fill the kernel

Tell the AI what your business is.

The kernel is a JSON file. It has a fixed shape depending on your vertical. Fill the fields for your business.

Critical: the not_present list. This is what the AI will NEVER claim exists. If you don't have a hot tub, list "hot tub". If you don't service commercial kitchens, list them. This is what stops hallucinations.

15 minutes typical

Most businesses can fill their initial kernel in 15 minutes. You can refine it endlessly later — every real customer message that comes in is an opportunity to sharpen the kernel.

If you want to speed this up, upload your existing website, PDF menu, or past reply exports and let the AI extract a kernel draft. You review and finalise.

Step 4 · BYOK

Bring your own AI key.

Open autopilot.html on your deployed site. Pick a provider:

Anthropic

Claude · best quality. Get key at console.anthropic.com.

OpenAI

GPT · widely available. Get key at platform.openai.com.

Google Gemini

Gemini Pro. Get key at aistudio.google.com.

Groq

Fastest inference · llama-3.3-70b. Get key at console.groq.com.

Mistral

EU-hosted. Get key at console.mistral.ai.

WebLLM (free)

Llama 3.1 8B in your browser. Zero cost forever. Slower but no key needed.

Paste the key. It stores in your browser localStorage only. Nothing ever hits our servers.

Step 5 · Harvest

Enrich the kernel from your online footprint.

Click Harvest on the autopilot page. The system pulls review URLs you provide (Airbnb, Booking.com, Google, Trustpilot), competitor pages, local weather, seasonality signals. All are digested via your BYOK LLM into an enriched kernel.

This is what makes the AI sound like you, not like a generic chatbot. It learns your voice from your past replies. It learns your positioning from your listing copy. It learns what your customers actually complain about from your review distillation.

Step 6 · Practice

Train on the simulator before real customers.

Open sim.html. Click DOWNLOAD & TRAIN. Botler (Llama 3.1 8B) downloads to your browser once (~4GB, forever cached). Then it becomes a swarm of believable fake customers.

Set intensity (chill / normal / peak / chaos), duration, click Start Wave. Fake customers flood the inbox. You reply. Botler judges you 0-10 on warmth, accuracy, completeness, tone-fit. Practice until you're fast and confident.

Sim state is namespaced separately. Reset wipes only sim data. Nothing real is touched.

Step 7 · Slide the dial

Ratchet autonomy up as trust grows.

Watch (week 1)

AI narrates. Does nothing. You do it all. Pure learning.

Suggest (week 2-3)

AI drafts every reply. You approve.

Auto-Low (week 4-6)

FAQ replies auto-sent.

Auto-Med (month 2)

Quotes + pricing tweaks auto-sent.

Full (month 3+)

Everything auto except explicit flags.

Watch resume

Slide back down anytime.

Step 8 · Snap in modules

Install capabilities.

Open modules.html on your instance. Browse the marketplace. Click Install on any module. It downloads to your instance, registers its tools into the agent registry, adds any UI surfaces to your sidebar.

Or · done for you

Not the DIY type? Book Pro setup.

We deploy FallHub for you within 48 hours. Kernel pre-populated from your website + existing reviews. Autonomy calibrated. Modules installed. Autopilot tuned. You show up to a working autopilot on day one.

Reference

Docs · deep dives.