Repository navigation
Expand file tree
/
Copy path.env.example
More file actions
92 lines (77 loc) · 4.7 KB
/
Copy path.env.example
File metadata and controls
92 lines (77 loc) · 4.7 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
# Copy to .env and fill in. Never commit .env.
NEBIUS_API_KEY=
NEBIUS_BASE_URL=https://api.tokenfactory.nebius.com/v1/
# Model IDs as served by Token Factory; run scripts/00_list_models.py to see them.
# PLANNER is the live model (every decision and reply; pick for speed). NIGHTLY does the work nobody waits for
# (digest, memory tidy-up, trends, self-check; pick for quality). VLM looks through the camera. EMBED searches memory.
KIN_PLANNER_MODEL=nvidia/nemotron-3-super-120b-a12b
# Nemotron's reasoning mode for the planner. Keep it on: with it off Super stopped calling its tools (it promised
# a reminder without setting one) and was no faster.
KIN_PLANNER_THINKING=on
KIN_VLM_MODEL=google/gemma-3-27b-it
KIN_NIGHTLY_MODEL=nvidia/Nemotron-3-Ultra-550b-a55b
# Optional: run the night's thinking (memory tidy-up, weekly trends, self-check) in a Nebius serverless container
# instead of on the hub: the URL of `python -m kin.nightly --serve` and its bearer token. See docs/NEBIUS_JOB.md.
# Leave empty and the hub does it. If the container fails, the hub does it anyway and logs a nightly_error.
KIN_NIGHTLY_URL=
KIN_NIGHTLY_TOKEN=
KIN_EMBED_MODEL=Qwen/Qwen3-Embedding-8B
# Vector's voice stretch factor; above 1.0 is slower and easier to follow.
KIN_SPEECH_SLOWDOWN=1.2
# Optional: where the local memory lives (default data/kin.db). Do not put it in a cloud-synced folder.
KIN_DB_PATH=
# Optional: Vector serial, needed only if more than one robot is in the SDK config.
VECTOR_SERIAL=
# Optional: Tavily web search (https://tavily.com). Without a key Kin has no web_search tool.
TAVILY_API_KEY=
# Optional: what Kin calls the person, e.g. "Anna". Used in greetings, family messages and the daily note.
KIN_PERSON_NAME=
# Optional: the person's town, e.g. "Springfield, Illinois". Used for weather and local questions.
KIN_LOCATION=
# 'metric' or 'imperial': the units Kin speaks in (weather).
KIN_UNITS=metric
# How Kin speaks: 'vector' (the robot's own voice) or 'hub' (natural offline voice through the hub's speaker).
KIN_MOUTH=vector
# How Kin listens: 'console' (typed replies, for development) or 'hub' (the hub's microphone, offline Whisper).
KIN_EARS=console
# Hub voice, by short name. Female: lessac, amy, kristin, hfc_female, cori, jenny, alba. Male: ryan, ryan_fast, joe,`n# hfc_male, john, bryce, norman, kusal, alan, northern. Or any of Piper's voices by full name (scripts/voices.py catalog).
# List them and hear samples with: python scripts/voices.py
KIN_VOICE=lessac
# Seconds of silence that mean the person has finished speaking. Lower is snappier; 1.2 cut people off.
KIN_END_PAUSE=1.5
# Optional audio devices for hub mode: an index or part of a name, e.g. "Jabra". Empty = system default.
KIN_AUDIO_INPUT=
KIN_AUDIO_OUTPUT=
# Time of the daily check-in, 24-hour HH:MM.
KIN_CHECKIN_TIME=09:00
# Optional: the family side. Create a bot with @BotFather in Telegram and paste its token here.
TELEGRAM_BOT_TOKEN=
# Telegram chats allowed to talk to Kin, comma-separated. Message the bot once and it tells you your chat id.
KIN_FAMILY_CHAT_IDS=
# When the daily note is sent to the family, 24-hour HH:MM.
KIN_DIGEST_TIME=20:00
# Weekday (mon..sun) on which the family also gets the weekly note; 'off' disables.
KIN_WEEKLY_DAY=sun
# Reminders: minutes between tries, and tries per day before the family is told.
KIN_RENUDGE_MINUTES=10
KIN_REMINDER_ATTEMPTS=3
# Presence: from START Kin quietly looks for the person now and then; if there is no sign of them by
# DEADLINE he calls out once and then tells the family. DEADLINE=off disables it.
KIN_PRESENCE_START=07:30
KIN_PRESENCE_DEADLINE=10:30
KIN_PRESENCE_LOOK_MINUTES=30
# on: Kin never drives Vector off his charger; he looks, talks and listens from it. Docking is his least reliable
# move, and a robot that cannot get back runs flat and goes dark. off: he drives to the person (demos, filming).
KIN_STAY_ON_CHARGER=on
# on: between tasks Kin keeps a light hold on Vector so his own idle behaviours do not drive him off the charger to
# explore (he ran flat that way). "Hey Vector" and his button still work; he just does less on his own. off: he roams.
KIN_IDLE_HOLD=on
# "Hey Vector" and his back button: wire-pod keeps his built-in commands, and hands anything else to Kin through a
# local gateway (switch it on with: python scripts/wirepod_link.py --link). Vector acknowledges in his own voice with
# one of these phrases ('|' separated; empty = the built-in list). Port 0 turns the gateway off.
KIN_GATEWAY_PORT=8765
KIN_ACK_PHRASES=
KIN_WIREPOD_URL=http://localhost:2427
# off: you talk to Kin with "Hey Vector" or his back button; a pet is just a pet (he purrs, and it still counts as a
# sign you are around). on: petting his back also starts a conversation, heard through the hub microphone.
KIN_PET_TO_TALK=off