Why the fair-use limits exist
Every SXO agency plan includes two AI fair-use counters: copilot queries per day and AI report runs per month. These are not arbitrary — they exist because a single careless automation loop across client sites can spend more on language model tokens in an afternoon than an entire month of subscription fees.
We keep the limits out of the way for normal agency use. A delivery team working with copilots all day will rarely hit the Foothold cap of 100 queries. The limits only bite when something is wrong: an infinite retry loop, a pasted wall of context that should have been a link, or a custom agent that forgot to stop.
Think of the counters as a circuit breaker for your AI spend. When they trip, it is almost always because there is a specific fix to make — not because your plan is too small.
What counts against what
A copilot query is a single chat turn — one question in, one answer back. Typing in the client or agency copilot, asking the dashboard to summarise a sprint, or clicking a one-shot action button all count as one query each.
An AI report is a longer multi-step run that produces a finished document: a financial summary, a competitive edge analysis, a SWOT, a client intake digest. Reports are where the real token spend lives, so they get their own monthly budget.
Background automations — webhook triggers, scheduled syncs, cron-driven cleanups — count against copilot queries if they call a language model, otherwise they are free. Most built-in automations are rule-based and do not touch the AI at all.
Keep sessions short and focused
Long, sprawling chat threads are the single biggest source of runaway AI spend. Every new message in a long thread re-reads all the prior messages, which means the tenth question can cost ten times as much as the first.
Start a new session whenever you change topics. If you have been debugging a deployment and now want to draft a marketing email, open a fresh chat. Your copilot does not lose context by starting over — it gains focus.
Clear finished threads at the end of the day. A good rule of thumb: if you would not want a new teammate to read the full transcript to catch up, the thread has outlived its usefulness.
Write prompts that do more per query
A single well-scoped prompt beats ten exploratory ones. Before sending a question, ask yourself: what is the exact outcome I want, and what would a correct answer look like?
State the output format up front. "Give me a bullet list of three risks, each under 20 words" burns one query. Asking the same question three times with increasing specificity burns three.
Attach links, not walls of text. If you need the copilot to reference a document, paste the internal link — our copilots can read your vault, wiki, agency notes, and approved client notes directly. Pasting the full document as context is usually waste.
Batch related questions into one turn. "Summarise this meeting, extract action items, and assign owners" is one query. Three separate queries asking each piece cost three.
Save AI reports for when a copilot turn will not do
AI reports are powerful but expensive. Use them when you need something polished, structured, and citable — board memos, client deliverables, monthly financials, comp-edge analyses.
Do not run an AI report to answer a quick question. A copilot turn is fifty times cheaper and usually just as accurate for one-off lookups.
If you are iterating on a report (first draft, second draft, final), stay inside the existing report rather than running it from scratch each time. Re-runs share the cached context; cold starts do not.
Schedule recurring reports off-peak. If your team runs the same weekly revenue digest every Monday at 9am, move it to Sunday night — you will never compete with a live meeting, and you free up your fair-use budget for ad-hoc work during the week.
Audit your automations once a month
Automations are the number one cause of surprise quota hits. An agent that was supposed to retry three times keeps retrying. A webhook that was supposed to fire on new invoices fires on every invoice update, including its own. A draft-generation cron was left running after the team moved to a manual workflow.
Pick a day each month — first Monday is a good default — and skim the automation log for anything that looks off: runs without a clear source, spikes in frequency, errors that repeat. Turn off anything that is not earning its keep.
If you are unsure whether an automation is worth its AI cost, switch it off for a week. If nobody notices, it was not.
Staying under your limits
Foothold includes 100 copilot queries per day and 20 AI reports per month — plenty for an emerging agency running up to 10 client sites on Sx8s. If you are routinely hitting those numbers, your agency is ready for Stronghold.
Stronghold includes 1,000 copilot queries per day and 100 AI reports per month. A scaling agency using the platform across up to 25 client sites rarely crosses that line unless something is misconfigured.
Dominion and Apex are tuned for larger client portfolios and come with implementation support to tune model routing, client-site automations, and pay-per-use overrides to your usage pattern.
If a fair-use counter trips, the copilot pauses rather than silently burning your budget. You get a clear prompt to either upgrade, add client-site capacity, or fix the automation that tripped it.