Configuration

Every setting in areao1.yaml, the file at the top of your workspace, with its default.

areao1.yaml holds settings only. Secrets never go in it: tokens, webhook URLs and passwords live in the OS keychain, and the file only names the keychain entry (secret_ref). Many settings also have a control in Settings in the dashboard, which writes this same file.

A new workspace gets a complete areao1.yaml with every default written out. You can delete any key to go back to its default. Keys are checked when the file loads, so a typo or an unknown key is reported instead of ignored. The JSON Schema is in your workspace at .areao1/schemas/areao1-config.schema.json, and the models are in areao1/core/models.py.

Top level

Key	Default	What it does
schema_version	1	File format version. Don't change it.
workspace_name	the folder name	A name for this case
profile	o1a	Active criteria profile: o1a or eb1a. The header's profile switch changes it.
engine	openai	The agent engine. OpenAI is the only one.
overrides	{}	Your status overrides per profile: {profile_id: {criterion_id: dropped \| gap}}

server

Key	Default	What it does
server.host	127.0.0.1	Always 127.0.0.1; no other value is accepted
server.port	7777	Port for areao1 up (1024 to 65535). --port overrides it.

agent

The agent runs on OpenAI and only when you've added an API key (Settings > Your AI) or set OPENAI_API_KEY. See Add an OpenAI API key and Costs and budget caps.

agent:
  models:
    hard: gpt-6.1-sol      # chat, tasks you start, letters ($2 / $10 per 1M tokens in / out)
    mid: gpt-6.1-sol       # missions, the rule check, PDFs, cheap-mode chat
    mundane: gpt-6-luna    # chat extraction, long-chat summaries, mail sorting ($0.10 / $0.50)
  effort: medium
  web_search: true
  budget: {per_run_tokens: 300000, per_run_usd: 2.0, monthly_tokens: 10000000, monthly_usd: 50.0}

Key	Default	What it does
agent.models.hard	gpt-6.1-sol	Chat, runs you start by hand, letter drafts
agent.models.mid	gpt-6.1-sol	Missions, the rule-check judge, PDFs, cheap-mode chat
agent.models.mundane	gpt-6-luna	Chat extraction, long-chat summaries, mail sorting
agent.effort	medium	Reasoning effort: low, medium, high, xhigh or max
agent.cheap_mode	false	Run chat on the mid tier instead of the hard one
agent.web_search	true	Let the agent search the web
agent.max_searches	8	Web searches per run (0 to 50). At the cap the run finishes with what it found and says so.
agent.max_turns	25	Most tool-use turns per run (1 to 200)
agent.budget.per_run_tokens	300000	Token cap per run
agent.budget.per_run_usd	2.0	Dollar cap per run (null for none)
agent.budget.monthly_tokens	10000000	Token cap per calendar month
agent.budget.monthly_usd	50.0	Dollar cap per calendar month (null for none)
agent.missions.opportunity_scout	false	Weekly mission: find opportunities for your weakest criteria
agent.missions.what_changed	false	Daily mission: review what changed and write the briefing
agent.autopilot.tracker_updates	false	Apply pipeline, letter and deadline edits without asking
agent.autopilot.metrics	false	Apply metric values the agent read from a page and quoted
agent.autopilot.tier1_deadlines	false	Apply new deadlines quoted from a Tier 1 source (primary law or agency pages)

Tokens counted toward the budget are input, output and cache-creation tokens (cache reads are recorded, not counted). Dollar amounts come from Area O1's built-in price table. When the monthly cap is reached, new runs are refused until next month.

Autopilot can never apply anything that could affect a criterion (evidence, exhibits, overrides, the profile), whatever you set here. The environment variables AREAO1_MODEL_HARD, AREAO1_MODEL_MID and AREAO1_MODEL_MUNDANE (or the same names in the workspace .env) override the model lines.

privacy

Key	Default	What it does
privacy.redact_before_llm	true	Remove emails, phone numbers and currency amounts from text sent to the model

notifications

Desktop notifications work out of the box. To add email, Slack, Discord or ntfy, add a channel under notifications.channels and route events to it:

notifications:
  channels:
    desktop: {kind: desktop}
    slack: {kind: slack, secret_ref: notify:slack}   # detail defaults to full; use minimal for counts only
    phone: {kind: ntfy, topic: pick-an-unguessable-topic, detail: minimal}
  routes:
    deadline: [desktop, phone]
    digest: [slack]

Store the secret in the keychain, never in the file: areao1 secret set notify:slack. Then run areao1 notify test. Anything sent to Slack, Discord, email or ntfy leaves your computer, so detail: minimal sends only counts ("2 deadlines this week"), never titles.

Channels

Each entry under notifications.channels is a named channel:

Key	Default	Used by	What it does
kind	required	all	desktop, email, slack, discord or ntfy
enabled	true	all	Turn the channel off without deleting it
detail	full	all	minimal sends counts only, no titles
secret_ref	none	email, slack, discord, ntfy	Keychain entry: the webhook URL, SMTP password or ntfy access token
host	none	email	SMTP server
port	587	email	SMTP port (465 uses SSL)
starttls	true	email	Use STARTTLS (ignored on port 465)
username	none	email	SMTP login; the password is the secret_ref
from_addr, to_addr	none	email	Sender and recipient
server	https://ntfy.sh	ntfy	ntfy server, or your own
topic	none	ntfy	Topic name; pick an unguessable one

Routes

notifications.routes maps each event to a list of channel names. Every event goes to desktop by default.

Event	Sent when
deadline	A deadline is near or overdue
digest	The weekly digest is ready
new_candidates	A sync found new items for your Inbox
sync_error	A source failed to sync
mission	An agent mission finished
vault	An official source changed
opportunity	The daily opportunity check found something
test	areao1 notify test

Key	Default	What it does
notifications.deadline_alert_days	[14, 3, 1, 0]	Alert when a deadline is this many days away

schedules

A cron expression per job. Set a job to '' to turn it off. Jobs added in later versions get their default schedule until you set one.

schedules:
  sync: 0 8 * * *
  metrics-snapshot: 0 9 * * mon     # weekly; skips unless 13+ days since the last snapshot
  deadline-check: 0 7 * * *
  digest: 0 17 * * fri
  mission-opportunity-scout: 0 9 * * fri
  mission-what-changed: 0 7 * * *
  vault-watch: 0 6 * * *
  google: '*/15 * * * *'
  daily-opportunities: 30 8 * * *

The missions and daily-opportunities do nothing until they're turned on (agent.missions, opportunities.enabled). See Scheduled jobs.

vault

Key	Default	What it does
vault.enabled	true	Fetch the public official sources listed in the vault manifest
vault.community	true	Pull hash-verified snapshots of blocked official pages from the community library
vault.community_url	https://raw.githubusercontent.com/ris3abh/areao1-community-vault/main	Where the community library's manifest.json lives
vault.share_captures	false	Share captures of public government pages with the community library: written to a local outbox, and pushed (with gh, when you can push to the library) by vault-watch's check of the standard when a quoted source's wording changed or the library's copy is over 30 days old
vault.report_wording	false	When vault-watch finds that a source Area O1 quotes changed its wording, open a GitHub issue in vault.report_repo with gh (your GitHub login)
vault.report_repo	ris3abh/areao1	Where those issues go (owner/repo)

mail

Key	Default	What it does
mail.model_sorting	false	Send mail no rule could sort to the mundane model (sender, subject, first lines, redacted). Off: rules only.

outreach

Key	Default	What it does
outreach.daily_limit	10	Most emails sent per day (0 to 100), checked when you approve
outreach.follow_up_days	7	Quiet days after your email before a follow-up draft (2 to 60)

See Outreach etiquette.

opportunities

Key	Default	What it does
opportunities.enabled	false	Run the daily opportunity check every 24 hours (Settings > Gmail)
opportunities.verify_on_web	true	Confirm each event on its official page (or Devpost / MLH); may run one cheap web search per find with no usable link

See Daily opportunity check.

connectors

Key	Default	What it does
connectors.include_forks	false	Include forked GitHub repos
connectors.min_stars_for_candidate	5	Star threshold for proposing a GitHub repo as a candidate
connectors.min_downloads_for_candidate	100	Download threshold for proposing a Hugging Face model or dataset as a candidate

Calendar subscription

Deadlines and pipeline follow-ups are written to data/calendar.ics on every change. There's nothing to configure:

• To subscribe from Apple Calendar, Outlook or Thunderbird on the same computer, use webcal://127.0.0.1:7777/calendar.ics while areao1 up is running (use your port if you changed it).
• Or import the data/calendar.ics file.
• Google Calendar can't reach your laptop's localhost, so import the .ics file there.