Configuration

Files

Path	Holds
~/.config/narratty/config.toml	Your defaults and limits (below)
~/.config/narratty/lexicon.toml	Your pronunciation entries
~/.config/narratty/approvals.json	Sandbox permissions you approved, per spec
~/.local/share/narratty/	Downloaded voices and models
~/.cache/narratty/	Audio cache, workspace snapshots, build caches

Paths follow the platform's conventions (on macOS under ~/Library).

config.toml

[end_card]
enabled = false                    # no closing card unless a spec or flag asks for it

[sandbox]
max_network = "allowlist"          # none | allowlist | full
allow_env = ["LM_LICENSE_FILE"]    # env_passthrough names a spec may use
allow_mounts = true
allow_ssh_agent = false
allow_docker = false               # no spec gets your Docker/Podman engine
allow_environment = ["image", "compose", "container"]
allow_packages = true              # environment.packages and setup

Every key is optional. A command-line flag beats the spec, and the spec beats this file. [sandbox] is a cap: a spec asking for more fails. See Sandboxed runs.

Environment variables

Variable	Effect
NARRATTY_RUNTIME	auto, native, docker or podman (same as --runtime)
NARRATTY_IMAGE	Container image (same as --image)
NARRATTY_OFFLINE	1: never download voices (same as --offline)
NARRATTY_YES	1: approve sandbox permissions without asking (same as --yes)
NARRATTY_CONFIG_DIR	Directory of config.toml and approvals.json
NARRATTY_DATA_DIR	Directory for voices and models
NARRATTY_CACHE_DIR	Directory for caches
NARRATTY_AGENT_DIR	Local narratty-agent builds (<dir>/<arch>/narratty-agent) instead of the image's
NO_COLOR	Plain output