Add an OpenAI API key

An OpenAI API key turns on Area O1's chat and web lookups; everything else works without one.

Add the key

1. Create a key at platform.openai.com > API keys. OpenAI keys start with sk-.
2. In Area O1, open Settings > Your AI (or the Your AI step of onboarding).
3. Paste the key and press Check and save.

Area O1 checks the key with OpenAI's free model-list request before saving it. If it works, it's stored in your OS keychain (entry openai:api_key), never in your workspace. The card then says Using your OpenAI API key.

To remove it, press Forget the key on the same card.

Or use an environment variable

If OPENAI_API_KEY is set in the environment Area O1 starts from, it's used when no key is saved in Settings:

export OPENAI_API_KEY=sk-...
areao1

A key saved in Settings wins over the environment variable.

If the key isn't accepted

Area O1 says	What to do
That doesn't look like an OpenAI API key	Paste the whole key; it starts with sk-
That's an Anthropic key	Area O1 runs on OpenAI only; create an OpenAI key
OpenAI refused that key	Check you copied all of it and that it isn't revoked
Couldn't reach OpenAI to check the key	Check your connection and try again

What needs a key, and what doesn't

Needs a key	Works without one
Ask, the chat panel on every page	Onboarding from your LinkedIn PDF
Agent runs you start, and missions (scheduled agent runs)	Every connector: GitHub, Hugging Face, Semantic Scholar, OpenAlex, arXiv, ORCID, websites
Onboarding's web search for an award, judging role or membership	The Inbox, Evidence, Metrics, Pipeline, Contacts, Calendar, Knowledge and Memory pages
Draft from claims on the Letters page	The criteria scoreboard and DASHBOARD.md
The rule check against official sources	Gmail with rules-only mail sorting, .eml drops and sender checks
Reading a PDF that isn't a LinkedIn export	Chat-history imports with the local rules
People, asks, decisions and metrics from chat imports	The knowledge vault and its change signals
Model sorting of mail no rule could sort (off by default)	Notifications, the calendar file and scheduled jobs

Without a key, the features in the left column say so and do nothing else. Nothing breaks.

Claude Code and Codex don't need this key

If you use Area O1 from Claude Code, Codex or another MCP client, that agent runs on its own account. The MCP server needs no OpenAI key. See Set up your agent.

The three model tiers

Area O1 picks a model by task from a fixed table, never by asking a model. There are three tiers, one line each in agent.models in your workspace's areao1.yaml:

agent:
  models:
    hard: gpt-6.1-sol      # chat, tasks you start, letters
    mid: gpt-6.1-sol       # missions, the rule check, PDFs, cheap-mode chat
    mundane: gpt-6-luna    # chat extraction, long-chat summaries, mail sorting

Tier	Default model	Used for
hard	gpt-6.1-sol	Chat, runs you start by hand, letter drafts
mid	gpt-6.1-sol	Missions, the rule check, reading non-LinkedIn PDFs, chat in cheap mode
mundane	gpt-6-luna	Chat-import extraction, summaries, mail sorting, each web search

To change a tier, change its line. An environment variable (or a line in the workspace's .env) overrides the yaml: AREAO1_MODEL_HARD, AREAO1_MODEL_MID or AREAO1_MODEL_MUNDANE.

A model that isn't in Area O1's price table still runs: its tokens are counted and the token caps apply, but its cost shows as unknown.

The Agent page shows which model each tier uses.

Other agent settings

These also live under agent: in areao1.yaml:

Setting	Default	What it does
effort	medium	How hard the model reasons: low, medium, high, xhigh or max
web_search	true	Whether runs may search the web
max_searches	8	Web searches per run; at the cap the run finishes with what it found and says so
max_turns	25	Model turns per run
cheap_mode	false	Chat on the mid tier instead of hard
budget	see Costs	Per-run and monthly caps

Cheap mode

Settings > Your AI > Cheap mode runs chat answers on the mid tier instead of the hard one. Missions, checks and summaries are unchanged. With the shipped defaults both tiers use the same model, so cheap mode only saves money once you point hard at a larger model.

For prices and caps, see Costs and budget caps.