Installation

Requirement: Go 1.24+ and at least one reachable Ollama server.

git clone https://gitlab.techeve.de/techeve/ollama-agent.git
cd ollama-agent
make build                 # produces bin/ollama-agent

Quick start

./bin/ollama-agent config init > config.json   # generate example config
# adjust config.json: endpoints + models to your own environment
./bin/ollama-agent config validate             # validate config
./bin/ollama-agent endpoints                   # are the servers reachable?
./bin/ollama-agent run "Why is the sky blue?"

The config is searched for in this order: -config <path> → $OLLAMA_AGENT_CONFIG → ./config.json → ~/.ollama-agent/config.json.

Docker

The daemon (including the web UI) runs as a container — configuration can come entirely from environment variables:

docker build -t ollama-agent .
docker run -d --name ollama-agent -p 2330:2330 \
  -v ollama-agent-data:/data \
  -e OLLAMA_URLS=http://192.168.6.80:11434,http://192.168.6.80:11436 \
  -e OLLAMA_DEFAULT_MODEL=qwen3.5:9b \
  ollama-agent

Env variable	Meaning
OLLAMA_URLS	Comma list of Ollama servers (required without config.json); first URL = highest priority
OLLAMA_DEFAULT_MODEL	Default model (required without config.json)
OLLAMA_AGENT_ADDR	Bind address, default 0.0.0.0:2330
OLLAMA_AGENT_TOKEN	optional bearer token for API/UI
OLLAMA_AGENT_STATE	State directory, default /data (sessions, matrix, jobs)

For task routing, skills, MCP, and cron jobs, mount a complete config.json at /data/config.json — it always wins over the environment variables. Template: docker-compose.example.yaml.