Command Line

ollama-agent [-config path] <command>

Command	Function
run [flags] "prompt"	one-off agent run; flags: --category, --model, --skill, --json, --no-stream
chat [--model m] [--category c] [--skill s]	interactive chat; meta-commands: /model <name|auto>, /reset, /quit
serve [-addr host:port]	daemon: HTTP API + web UI + scheduler + MCP connections
ui	like serve, additionally opens the web interface in the browser
endpoints	pings all servers, status + model count (exit code 1 if one is down — handy for monitoring)
models	models per server + task mapping with availability
skills [list|show <name>]	list / show skills
tools	all tools (built-in + MCP) and MCP server status
jobs [list|run <id>|history <id>]	manage scheduler jobs
probe [--models m1,m2] [--exclude p1,p2] [--ctx-max n]	standardized capability test of all models → model matrix
probe deep [--plan] [--ctx n]	context depth test: how well does a model actually use its large context (percentage score)
probe show	show saved model matrix
config [validate|init]	validate config / print example
version	show version

Examples:

# Explicit model instead of routing
./bin/ollama-agent run --model qwen3-coder:latest "Refactore diese Funktion: ..."

# Result as JSON (with category, model, tool calls, token counters)
./bin/ollama-agent run --json "..." | jq .

# Force a skill
./bin/ollama-agent run --skill changelog-writer "v0.2.0: Feature X, Bugfix Y"