Skip to content

Settings files

Both files are YAML and reload when you save them. Most settings can also be changed on your parent page.

The household’s settings, in the root of the household folder.

parent:
name: David
title: Dad # what your children call you
pin_hash: … # your PIN, hashed (argon2id)
models: # provider/model
tutor: anthropic/claude-sonnet-5-5
check: anthropic/claude-haiku-5-5
port: 4747
budget:
daily_tokens_per_child: 400000
session_idle_minutes: 30 # a new sitting starts after this long without a message
apps:
catalogues: [] # https links to app catalogues
Setting Default What it does
models.tutor chosen in init The model that talks with your child, as provider/model: anthropic/claude-sonnet-5-5, openai/gpt-5.5, openrouter/google/gemini-3.5-flash…
models.check chosen in init The smaller model used for the safety checks: anthropic/claude-haiku-5-5, openai/gpt-5.4-mini…

The two models can come from different providers. Each provider needs its key in .env (see Environment variables). OpenBuddies is tuned for the Anthropic and OpenAI models above. With any other model, openbuddies status shows a warning: the safety checks depend on the check model, so try it with the evals in a copy of the OpenBuddies repo (pnpm eval -- --tutor <model> --check <model>) before your child uses it. | port | 4747 | The port the server listens on. | | budget.daily_tokens_per_child | 400000 | Each child’s daily allowance, unless their rules.yaml sets one. | | session_idle_minutes | 30 | Minutes without a message before a new sitting starts. | | apps.catalogues | none | App catalogues you’ve added. |

One per child, in children/<child>/rules.yaml.

default_mode: strict
subjects:
history:
mode: relaxed
hint_ladder: [ask, nudge, worked_example, check]
max_leak_retries: 2
alerts:
welfare:
sensitivity: high
hours:
weekdays: "16:00-19:30"
weekends: "09:00-18:00"
daily_tokens: 300000
web:
mode: trusted
sites: [bbc.co.uk]
blocked: []
daily_lookups: 30
apps:
spelling-bee: on
pictures:
daily: 6
heartbeat:
mode: on
at: "16:30"
Setting Default What it does
default_mode chosen in init The child’s help level: strict, balanced or relaxed.
subjects.<subject>.mode the child’s level A help level for one subject. Subjects not listed use default_mode.
hint_ladder [ask, nudge, worked_example, check] The steps of help, in order.
max_leak_retries 2 How many times a reply that gives the answer away is rewritten before a safe fallback is sent.
alerts.welfare.sensitivity high How readily worrying messages alert you: high, medium or low.
hours.weekdays / hours.weekends any time When the child can use OpenBuddies, as HH:MM-HH:MM.
daily_tokens the household budget This child’s daily AI allowance. At least 10,000.
web.mode open Looking things up: off, trusted or open.
web.sites none Extra trusted sites.
web.blocked none Sites that are never used.
web.daily_lookups 30 Lookups a day.
apps.<app> off for added apps Turns an added app on or off for this child. Built-in apps are on.
pictures.daily 6 Pictures a day in Picture Studio.
heartbeat.mode / heartbeat.at on, 16:30 The daily check-in, when a buddy may say something useful (a test tomorrow, say).