Configuration

regit reads an optional YAML config file that seeds the TUI's forms and provides defaults for the headless CLI. Every value has a built-in default, so the file is entirely optional.

File location

$XDG_CONFIG_HOME/regit/config.yaml

falling back to:

~/.config/regit/config.yaml

A missing file is fine — regit runs on built-in defaults.

Precedence

Values are resolved lowest to highest precedence:

1. Built-in defaults
2. YAML config file
3. REGIT_* environment variables (env wins)

Loading is strict

A typo'd key or unusable value is a hard error at startup, not a silent fallback: unknown YAML keys, an unrecognized timezone, and an inverted workday window (workday_end not after workday_start) all abort with a clear message. This keeps a misspelled key from silently reverting a setting to its default.

Top-level keys

Key	Type	Default	Env var	Description
timezone	string	local	REGIT_TIMEZONE	IANA zone name (e.g. Europe/Berlin) or local for the machine zone. Used when computing new commit dates.
workday_start	HH:MM	09:00	REGIT_WORKDAY_START	Start of the default working-hours window.
workday_end	HH:MM	18:00	REGIT_WORKDAY_END	End of the working-hours window. Must be after workday_start.
lunch_start	HH:MM	(empty)	—	Start of an optional lunch gap that time models avoid. Empty disables the gap.
lunch_end	HH:MM	(empty)	—	End of the lunch gap.
default_strategy	string	Even Spread	REGIT_DEFAULT_STRATEGY	Day-assignment strategy pre-selected when the TUI launches.
default_time_model	string	Working hours	REGIT_DEFAULT_TIME_MODEL	Time-of-day model pre-selected when the TUI launches.
jitter_minutes	int	20	—	Default ± spread (in minutes) for the Jitter time model.
link_dates	bool	true	REGIT_LINK_DATES	Keep author and committer dates equal by default.
weekdays	list of strings	[mon, tue, wed, thu, fri]	—	Enabled weekday short names (sun…sat) for weekday-masked strategies.
seed	int64	0	REGIT_SEED	Seed for randomized strategies, making them reproducible. 0 means "derive from the current time".

llm section

Configures the optional language model used by split, reword, and squash. Off unless a provider/model is set here or --llm/--agent is passed. See LLM providers for the full provider and credential story.

Key	Type	Default	Env var	Description
llm.provider	string	(empty)	REGIT_LLM_PROVIDER	Default provider (ollama, lmstudio, openai, openrouter).
llm.model	string	(empty)	REGIT_LLM_MODEL	Default model name for the provider.
llm.base_url	string	(empty)	—	Override the provider's default OpenAI-compatible endpoint.
llm.temperature	float	0	—	Accepted but currently unused — chat requests always send temperature 0 for deterministic output.
llm.timeout_sec	int	60	—	Per-request timeout in seconds for HTTP backends.

API keys are not config

There is no API-key config field, by design. Keys come only from environment variables (REGIT_LLM_API_KEY, OPENAI_API_KEY, OPENROUTER_API_KEY) — see LLM providers.

split section

Defaults for the LLM-backed commit-splitting feature.

Key	Type	Default	Env var	Description
split.max_commits	int	4	REGIT_SPLIT_MAX_COMMITS	Maximum commits one split can produce.
split.lines_per_commit	int	40	—	Target changed-line count per resulting commit, used to pick the split size.
split.min_split_lines	int	20	—	Minimum changed lines a commit needs before splitting it is considered.
split.recursion_depth	int	2	—	Accepted but currently unused by the split engine.
split.context	list of strings	[readme]	—	Extra context sources included in the prompt (e.g. readme).
split.repair_attempts	int	2	—	LLM repair-loop attempts when --test-cmd validation fails.
split.wip_prefix	string	"WIP: "	—	Subject prefix marking intermediate work-in-progress commits.
split.cache	bool	true	—	Accepted but currently unused — split proposals are always cached under .git/regit/cache/; pass --no-cache to bypass per run.

reword section

Defaults for the LLM-backed commit-rewording feature.

Key	Type	Default	Env var	Description
reword.variants	int	3	REGIT_REWORD_VARIANTS	Number of ranked message alternatives to generate per commit.
reword.context	list of strings	[readme]	—	Extra context sources included in the prompt (e.g. readme, filestates).
reword.context_lines	int	30	—	Diff context lines sent to the model.
reword.prefetch	int	1	—	Accepted but currently unused — the TUI always prefetches proposals one commit ahead.
reword.cache	bool	true	—	Accepted but currently unused — reword proposals are always cached under .git/regit/cache/; pass --no-cache to bypass per run.

Environment overrides

Every supported REGIT_* override (env beats the file; unparsable numbers and bools are ignored and fall back to the file/default):

• REGIT_TIMEZONE
• REGIT_WORKDAY_START
• REGIT_WORKDAY_END
• REGIT_DEFAULT_STRATEGY
• REGIT_DEFAULT_TIME_MODEL
• REGIT_SEED
• REGIT_LINK_DATES
• REGIT_LLM_PROVIDER
• REGIT_LLM_MODEL
• REGIT_SPLIT_MAX_COMMITS
• REGIT_REWORD_VARIANTS

Full example

A complete config.yaml showing every key at its built-in default:

timezone: local
workday_start: "09:00"
workday_end: "18:00"
lunch_start: ""
lunch_end: ""
default_strategy: Even Spread
default_time_model: Working hours
jitter_minutes: 20
link_dates: true
weekdays: [mon, tue, wed, thu, fri]
seed: 0

llm:
  provider: ""
  model: ""
  base_url: ""
  temperature: 0
  timeout_sec: 60

split:
  max_commits: 4
  lines_per_commit: 40
  min_split_lines: 20
  recursion_depth: 2
  context: [readme]
  repair_attempts: 2
  wip_prefix: "WIP: "
  cache: true

reword:
  variants: 3
  context: [readme]
  context_lines: 30
  prefetch: 1
  cache: true

Related pages

• LLM providers — provider setup, API keys, and the agent backend
• CLI reference — flags that override these defaults per invocation
• redate — the strategies and time models these defaults seed
• Getting started — first run without any configuration