Installation

Installation

Four routes. Pick one.

Route	Best for	Time
Docker	least setup	~5 min
Linux / macOS	day-to-day	~5 min
Windows	day-to-day	~5 min
make	anyone with make	~3 min

Everywhere: Python 3.10+ (or Docker), ~500 MB disk, and a model — local via Ollama or a cloud API key. No GPU required.

---

Docker

git clone https://github.com/ElmatadorZ/skynetclaw.git
cd skynetclaw
cp backend/settings.example.json backend/settings.json

docker compose up -d
docker compose exec ollama ollama pull llama3.1:8b     # first run only
curl http://127.0.0.1:8766/api/system/health

Two containers: the backend on 127.0.0.1:8766, Ollama on :11434. Models and the database live in named volumes and survive docker compose down.

docker compose down -v also deletes the volumes — that erases the House's memory. See Docker.

---

Linux / macOS

git clone https://github.com/ElmatadorZ/skynetclaw.git
cd skynetclaw
./start.sh

start.sh is idempotent: it creates .venv, installs the five dependencies, seeds both config templates, runs the migration, warns if Ollama is unreachable, and starts the server. Run it again whenever — it skips what is already done. A different port: ./start.sh 9000.

Distribution packages and discovery details: Linux.

---

Windows

git clone https://github.com/ElmatadorZ/skynetclaw.git
cd skynetclaw

copy .env.example .env
copy backend\settings.example.json backend\settings.json

python -m venv .venv
.venv\Scripts\activate
pip install -r backend\requirements.txt

ollama pull llama3.1:8b

cd backend
python migrate.py up
python -m uvicorn main:app --host 127.0.0.1 --port 8766

install.bat and start.bat wrap the same steps.

---

make

make setup     # venv + dependencies + config + database
make run
make health
make test      # 601 tests

---

Verifying the install

curl http://127.0.0.1:8766/api/system/health
# {"ok":true,"status":"YELLOW","summary":"13 green · 1 degraded", ...}

YELLOW is the expected verdict here, not a problem. Before you start a model runtime the ollama check reports degraded — the House is up and honest about what it cannot reach. Start Ollama (or configure a cloud connection) and the same call returns GREEN — all 14 checks pass.

The distinction is deliberate:

Verdict	Meaning	ok
GREEN	every subsystem reachable	true
YELLOW	running, something optional is absent or not yet generated	true
RED	the House itself is faulty — a file, module, or database is broken	false

Each check names itself in the response, so read checks[] before changing anything.

→ Next: First Run