🧠 Executive / Assistant — go live in one command

This stack is an executive assistant:

                         ┌──────────────── Caddy (auto-HTTPS) ────────────────┐
   you ────────────────▶ │  https://assistant.yourdomain.com                  │
   (chat / search)                            │                               │
                         └───────────────────┼───────────────────────────────┘
                                             ▼
                                     Khoj (second brain)
                                     │            ▲
                        your docs    │            │  answers + citations
                                     ▼            │
                             semantic search over your knowledge
                                     │
                    Postgres/pgvector · SearXNG · Terrarium

You get a personal AI assistant that chats with and semantically searches your own documents — notes, PDFs, org files, Markdown — behind real HTTPS on your domain, and it can run entirely on a local/private LLM.

Replaces: an executive assistant + a "chat with my notes" tool + scattered search
Time to live: ~30–60 minutes
You need: a server with a public IP, a domain, and Docker

Step 1 — Get a server

Any Ubuntu 22.04+ VPS with 2 vCPU / 4 GB RAM (minimum) works — Hetzner, DigitalOcean, Vultr, AWS Lightsail, etc. SSH in, then install Docker:

curl -fsSL https://get.docker.com | sh

Open ports 80 and 443 in your provider's firewall.

Step 2 — Point your domain at it

Create a DNS A record for assistant.yourdomain.com → your server's IP. Verify it resolves before continuing (HTTPS issuance depends on it):

dig +short assistant.yourdomain.com   # should print your server IP

Step 3 — Get the stack

git clone https://github.com/arunrajiah/ai-workforce.git
cd ai-workforce/departments/executive/deploy
cp .env.example .env

Step 4 — Fill in `.env`

Generate the secrets and edit the file:

echo "KHOJ_DJANGO_SECRET_KEY=$(openssl rand -hex 32)"
echo "POSTGRES_PASSWORD=$(openssl rand -hex 24)"
echo "KHOJ_ADMIN_PASSWORD=$(openssl rand -hex 24)"

In .env, set at minimum:

DOMAIN → assistant.yourdomain.com
ACME_EMAIL → your email (for Let's Encrypt)
KHOJ_DJANGO_SECRET_KEY, POSTGRES_PASSWORD, KHOJ_ADMIN_PASSWORD → the generated values
KHOJ_ADMIN_EMAIL → the email you'll use to log into Khoj's admin panel

Step 5 — Launch

Bring everything up:

docker compose up -d

Caddy will fetch a TLS certificate automatically. In ~1 minute, open https://assistant.yourdomain.com and sign in.

Watch it come up: docker compose logs -f server caddy

Step 6 — Point Khoj at a local/private LLM

Khoj configures its chat model in the admin panel — this is where you keep it provider-neutral and offline.

Open https://assistant.yourdomain.com/server/admin and log in with KHOJ_ADMIN_EMAIL / KHOJ_ADMIN_PASSWORD from your .env.
Add an AI Model API. Set its api url to any OpenAI-compatible endpoint. For a local/private model, run Ollama on the host and use http://host.docker.internal:11434/v1/.
Add a Chat Model entry, choose type OpenAI, and set the model name (e.g. llama3.1) to match what your endpoint serves.

Prefer to wire it up front instead of in the UI? Set LLM_BASE_URL and LLM_API_KEY in .env (they map to Khoj's OPENAI_BASE_URL / OPENAI_API_KEY) and docker compose up -d. For local Ollama, any non-empty LLM_API_KEY works.

Keep it private: pointing the base URL at a local Ollama means your documents and prompts never leave the server. No external API key required.

Step 7 — Add your documents

Give your assistant a knowledge base to search and chat over:

Web/desktop: in Khoj, open Settings and connect your files — upload Markdown, PDF, .org, and plain-text docs, or sync a folder.
Khoj indexes them into semantic search automatically; ask a question and it answers with citations back to the source.

Step 8 — Use it every day

Chat with your knowledge, run semantic searches, and let Khoj draft and recall across everything you've indexed:

Search — find anything across your notes and docs by meaning, not keywords.
Chat — ask questions and get answers grounded in your own material.
Clients — install the Khoj Obsidian/Emacs/desktop clients and point them at https://assistant.yourdomain.com.

You're live. 🎉

Operate it

Task	Command
View logs	`docker compose logs -f server caddy`
Update Khoj	`docker compose pull && docker compose up -d`
Restart the server	`docker compose up -d server`
Back up the database	`docker compose exec database pg_dump -U postgres khoj > backup.sql`
Stop everything	`docker compose down` (add `-v` to also wipe data)

Troubleshooting

No HTTPS certificate — DNS isn't pointing at the server yet, or ports 80/443 are blocked. Check docker compose logs caddy.
502 / blank page — the server is still booting (first run downloads embedding models). Give it a minute and check docker compose logs server.
No answers / model errors — confirm you added an AI Model API and Chat Model in the admin panel (Step 6), and that your LLM endpoint is reachable from the container. For a host Ollama, use http://host.docker.internal:11434/v1/.
Docs not searchable — re-open Khoj Settings and re-sync the files; watch docker compose logs -f server while indexing.

Swap a component

Prefer a different assistant or a lighter footprint? Every piece is replaceable — see the Executive / Assistant directory for alternatives you can run instead.

Verification note: service names, images, the 42110 web port, and the POSTGRES_* / KHOJ_* / KHOJ_TERRARIUM_URL / KHOJ_SEARXNG_URL env vars are taken from Khoj's official docker-compose.yml. Khoj configures the chat model primarily through its admin UI (/server/admin); the OPENAI_BASE_URL / OPENAI_API_KEY env vars are Khoj's documented way to preset an OpenAI-compatible (or local Ollama) endpoint. LLM_MODEL is provided for your convenience but is set on the Chat Model entry in the admin UI. This compose omits Khoj's optional --anonymous-mode flag so the admin login is enforced; add it back if you want an open, single-user instance.