Costs and budget caps

Area O1 itself is free; the AI features bill your OpenAI account, and caps in your settings stop runs before they go over.

What you pay for

Only runs that use your OpenAI key cost money: chat, agent runs and missions, letter drafts, the rule check, web lookups, and the optional model steps (chat-import extraction, model mail sorting, reading a non-LinkedIn PDF). The daily opportunity check, once you turn it on, may run one cheap web search per find that has no usable link (opportunities.verify_on_web). See what needs a key.

Area O1 computes each run's cost from the tokens OpenAI reports and a price table in the app (areao1/engine/openai_engine.py). Prices for the default models, in US dollars per 1 million tokens:

Model	Input	Cached input	Output	Default tier
gpt-6.1-sol	$2.00	$0.10	$10.00	hard, mid
gpt-6-luna	$0.10	$0.01	$0.50	mundane
gpt-6-astra	$10.00	$1.00	$50.00	(an upgrade option)

Cache writes cost 1.25 times the input rate. Prompt caching is automatic, so repeated instructions are billed at the cached rate.

Web search costs $0.01 per search ($10 per 1,000), plus the tokens of the small sub-request that runs it on the mundane tier. A run makes at most agent.max_searches searches (default 8); at the cap it finishes with what it found and says so.

Prices change

These are the prices in the app's table (recorded 2026-10-07). OpenAI bills you at its own current rates, so check OpenAI's pricing page and your OpenAI usage dashboard.

Typical costs

Examples from one setup on the default models; yours will vary with how long the conversation is and how much the run reads:

Run	About
A chat answer	$0.03 or less
An opportunity-scout mission with 10 web searches (before the default cap of 8)	$0.23, of which about $0.11 was searches

Budget caps

Spend is capped per run and per month. The caps live under agent.budget in your workspace's areao1.yaml:

agent:
  budget:
    per_run_tokens: 300000
    per_run_usd: 2.0
    monthly_tokens: 10000000
    monthly_usd: 50.0

Setting	Default	What happens at the cap
per_run_tokens	300,000	The run stops and says the per-run token budget was reached
per_run_usd	$2.00	The run stops between turns once its cost reaches the cap
monthly_tokens	10,000,000	New runs are refused for the rest of the month
monthly_usd	$50.00	New runs are refused for the rest of the month

Tokens counted toward the caps are input, output and cache writes; cache reads are recorded but not counted.

When the monthly cap is reached, a chat-history import still runs with the local rules only, so you don't lose the import.

Edit the numbers to suit you, and save the file. To be extra safe, also set a spending limit in your OpenAI account.

Where costs show

The Agent page shows:

• This month: dollars spent, tokens used against your monthly token cap (with a bar that turns red near the cap), your dollar cap, and the per-run caps.
• Every run: its tier, model and cost, including how many searches it made and what they cost.
• Each run's detail page: cost, searches, what it read and what it proposed.

Chat-import extraction and model mail sorting count toward the same month.

Missions are off until you turn them on in Settings > Missions, since they spend tokens without anyone pressing a button. The daily what-changed mission is skipped, at no cost, when nothing changed.

What costs nothing

These never call a model:

• Connectors: GitHub, Hugging Face, Semantic Scholar, OpenAlex, arXiv, ORCID and websites, plus metrics snapshots and sync.
• Onboarding from a LinkedIn PDF, read on your computer.
• Gmail with rules-only sorting (the default), .eml drops and sender checks.
• Chat-history imports with the local rules (the areao1 import command always works this way).
• The knowledge vault: fetching official sources, change signals, the community snapshot library, search.
• The scoreboard, the Inbox, the calendar, notifications and every other page.
• Using Area O1 from Claude Code or Codex over MCP: those agents bill their own accounts, not your OpenAI key.