Installation (self-host)

Self-host instructions. The hosted multi-tenant variant lives at humancensys.com — see Two modes.

Prerequisites

• Docker Desktop (or Docker Engine + Compose v2)
• Node.js 20+ for the web UI dev server
• An Anthropic API key — get one at console.anthropic.com
• (Optional) LlamaParse API key — document parsing, cloud.llamaindex.ai
• (Optional) Tavily API key — web search, tavily.com

Clone and configure

git clone https://github.com/Lizo-RoadTown/Make_Skills.git
cd Make_Skills/platform/deploy
cp .env.template .env

Open .env in your editor and fill in:

ANTHROPIC_API_KEY=sk-ant-...           # required
LLAMA_CLOUD_API_KEY=llx-...            # optional but recommended
TAVILY_API_KEY=ydc-sk-...              # optional

Postgres runs in trust mode (no password) since the port isn't exposed to the host. If you ever expose port 5432, set a real POSTGRES_PASSWORD.

Bring up the stack

docker compose up -d --build
docker compose logs -f api    # watch the agent build

You'll see:

INFO:api:Building deepagents agent...
INFO:api:Agent ready.
INFO: Uvicorn running on http://0.0.0.0:8001

Three services are now running:

LanceDB runs embedded inside the api container.

Run the web UI

In a separate terminal:

cd ../../web
npm install
npm run dev

Open http://localhost:3000. The chat UI talks to the agent at http://localhost:8001.

Verify

curl http://localhost:8001/healthz
# {"status":"ok"}

curl http://localhost:8001/memory/stats
# {"total":0}  (until you have a chat — recorder fills this)

Optional: get to humancensys.com from your laptop

If you want to reach your local stack from a different machine, see Remote access. Tailscale is the simplest path — works for "my desktop is at home, I'm on my laptop somewhere else."

Optional: cloud-host this stack on Render

render.yaml provisions the stack on Render. Walkthrough at platform/RENDER.md. Approximately $7/month.

Stop / restart / nuke

docker compose stop                # pause; data preserved
docker compose start               # resume
docker compose down                # remove containers; volumes preserved
docker compose down -v             # nuke volumes too (lose conversations + memory)

What's next

• Agents and skills — the mental model
• Memory — how the recorder + LanceDB work
• Architecture — the layered model + container topology