Configuration

LocalAlibiScan works without any configuration. To change the defaults, create ~/.localalibi/config.toml:

las config --init     # writes the file below (never overwrites)
las config            # shows the configuration in use and where it comes from

[ui]
# Interface language: "en" or "pt" (also LOCALALIBI_LANG or las --lang-ui).
language = "en"

[scan]
# Folders to skip in addition to the fixed ones (node_modules, .venv, venv, dist,
# build, .git, __pycache__, .next, target, site-packages, any virtualenv) and
# each project's .gitignore.
ignore = []
# Files larger than this (bytes) are listed but not read.
max_file_size = 1048576

[dashboard]
# How many levels to walk down looking for projects.
depth = 2

[ollama]
# Only localhost is accepted. AI drafting is optional.
url = "http://localhost:11434"
model = "qwen2.5-coder:7b"
timeout = 180

Precedence

Defaults < config.toml < environment variables < command-line options:

Variable	Overrides
LOCALALIBI_LANG	[ui] language (las --lang-ui overrides both)
LOCALALIBI_OLLAMA_MODEL	[ollama] model
LOCALALIBI_OLLAMA_URL	[ollama] url (must still be localhost)
LOCALALIBI_CONFIG	path of the configuration file itself

An invalid file stops every command with a message naming the file and the problem, instead of being silently ignored.

Choosing a model

Any model you have in Ollama works (ollama list). Small coding models such as qwen2.5-coder:7b are a good fit: the model never sees your code in las explain, only short excerpts in las ask, and every sentence it writes is checked by the validator.

ollama pull qwen2.5-coder:7b
las explain . --model qwen3:8b     # one-off