Config Reference

This page is the practical reference for the settings users change most often. For a conceptual walkthrough, start with Configuration.

Two homes:

• config.yaml (system-owned): daemon, auth, google, browser, timezone, cron. Read at startup; the daemon treats the file as read-only.
• Settings database (dashboard-managed): everything else. Edited in the dashboard, applied live. A legacy block in config.yaml is imported once at the next restart; after that the database wins.

config.yaml (system)

daemon

Key	Type	Default	Notes
port	number	3142	Dashboard + daemon port
data_dir	string	~/.jarvis	Root data directory
db_path	string	~/.jarvis/jarvis.db	SQLite vault path
brain_domain	string	unset	External origin (explicit scheme) embedded in enrollment tokens
listen	string	unset	unix:/abs/path.sock for unix-socket mode instead of TCP
drain_deadline_ms	number	75000	Graceful shutdown budget

auth

Key	Type	Default	Notes
insecure_open_access	boolean	false	Escape hatch: serve dashboard/API with no auth. Setup only.

The old auth.token no longer exists; access is JWT-only via jarvis enroll.

google

Key	Type	Default	Notes
client_id	string	unset	OAuth client for Gmail/Calendar
client_secret	string	unset	Account authorization itself happens in the dashboard

browser

Key	Type	Default	Notes
local	boolean	true	false on headless servers: browser actions route to sidecars

timezone

Key	Type	Default	Notes
timezone	string	system	IANA zone for cron and goal windows

cron

Key	Type	Default	Notes
morning	string	0 7 * * *	Morning cycle
evening	string	0 20 * * *	Evening cycle
hourly	string	37 * * * *	Hourly cycle

Environment overrides

JARVIS_PORT, JARVIS_HOME, JARVIS_BRAIN_DOMAIN, JARVIS_WAKE_ENGINE, JARVIS_REALTIME_VOICE, JARVIS_TELEMETRY, DO_NOT_TRACK. LLM configuration is intentionally not readable from env vars.

Settings Database (dashboard-managed)

The shapes below are what the settings API stores. Change them in Settings; they hot-apply.

user

Key	Type	Default	Notes
name	string	""	Your name for personalization

llm

Configured in Settings → LLM. API keys live in the encrypted keychain and are never returned by the API.

Key	Type	Default	Notes
providers	map	{}	Named entries: { kind, api_key or base_url }. Kinds: anthropic, openai, groq, gemini, ollama, openrouter, nvidia, openai_compatible, litellm
default	string	unset	Single-model mode: "<provider>:<model>"
tiers	map	{}	Multi-tier mode: conversation, high, medium, low
prompt_cache	boolean	true	Prompt caching, per provider

tts

Key	Type	Default	Notes
enabled	boolean	false	Voice output
provider	enum	edge	edge, elevenlabs, sarvam
voice	string	en-US-AriaNeural	Edge voice
rate, volume	string	+0%	Percentage strings
elevenlabs.*	object		api_key, voice_id, model, stability, similarity_boost
sarvam.*	object		api_key, model, language, speaker, sampling_rate

stt

Key	Type	Default	Notes
provider	enum	openai	openai, groq, local, sarvam
openai.*, groq.*	object		api_key, model
local.*	object		endpoint (default http://localhost:8080), model, server_type (whisper_cpp or openai_compatible)
sarvam.*	object		api_key, model, language

voice

Key	Type	Default	Notes
wake_engine	enum	openwakeword	openwakeword, webspeech, auto
realtime.enabled	boolean	false	GPT Realtime speech-to-speech mode
realtime.model	string	gpt-realtime-2
realtime.reasoning_effort	enum	low	minimal to xhigh
realtime.max_session_minutes	number	10
realtime.monthly_budget_usd	number	unset

channels

Key	Type	Notes
telegram	object	enabled, bot_token, allowed_users (numbers)
discord	object	enabled, bot_token, allowed_users (strings), guild_id

authority

Key	Type	Default	Notes
default_level	number	3	The 1 to 10 authority floor
governed_categories	list	send_email, send_message, make_payment	Actions converted to approval requests
overrides	list	[]	Per-action allow/deny, optionally per role
context_rules	list	[]	Time/tool/always rules
learning.enabled	boolean	true	Auto-approve suggestions
learning.suggest_threshold	number	5	Approvals before suggesting
emergency_state	enum	normal	normal, paused, killed

awareness

Key	Type	Default
enabled	boolean	true
capture_interval_ms	number	7000
min_change_threshold	number	0.02
cloud_vision_enabled	boolean	true
cloud_vision_cooldown_ms	number	30000
stuck_threshold_ms	number	120000
struggle_grace_ms	number	45000
struggle_cooldown_ms	number	90000
suggestion_rate_limit_ms	number	60000
overlay_autolaunch	boolean	true
retention.full_hours	number	1
retention.key_moment_hours	number	24

sites

Key	Type	Default
enabled	boolean	true
projects_dir	string	~/.jarvis/projects
port_range_start / port_range_end	number	4000 / 4999
max_concurrent_servers	number	3

heartbeat

Key	Type	Default	Notes
interval_minutes	number	15	Legacy; the periodic heartbeat loop was removed in 0.9 and these values are no longer used for scheduling
active_hours.start / .end	number	8 / 23
aggressiveness	enum	aggressive

personality

Key	Type	Default
core_traits	list	loyal, efficient, proactive, respectful, adaptive
assistant_name	string	Jarvis

telemetry

Key	Type	Default	Notes
enabled	boolean	true	Anonymous ping at startup and hourly. Also disabled by JARVIS_TELEMETRY=0 or DO_NOT_TRACK=1. The sidecar has a separate toggle.

active_role

Key	Type	Default
active_role	string	personal-assistant

goals, workflows

Managed through their rooms; see Goals and Workflows for their setting shapes.