Guide · CompanionGuard · Published 14 November 2026 · Reviewed 14 November 2026 · By the Ingenious AI team
Session and Daily Caps: How the Runtime Enforces Breaks for Minors
The short answer
On Pro and Team, the operator can configure a session cap and a daily cap for minor sessions. The runtime tracks elapsed session and daily minutes for each minor session reference; at a cap it returns enforce_break instead of allow, and the evidence event records the elapsed values that triggered it. The caps are entirely operator-configured — the product presents no cap value as a statutory number, because the packs set a reminder cadence (at least every 3 hours in California) but no session-length law.
What is a minor session cap?
A minor session cap is an operator-set limit — minutes per session, and minutes per day — that the Crisis Evidence Runtime enforces for sessions flagged as minor sessions. Reaching either cap changes the runtime's answer from allow to enforce_break, and the enforcement is written into the hash-chained evidence with the elapsed session and daily values that caused it.
Create the free safety kit Next guide All guides
CompanionGuard is built by Ingenious AI. A compliance toolkit — not legal advice and not a clinical service. If you are in crisis right now: US — call or text 988; UK & Ireland — Samaritans 116 123.
Caps are enforcement; reminders are offers
The minor-safety design has two gears that are easy to confuse. The free kit's break reminders are offers: at least every 3 hours, the snippet gently suggests a break to a known minor — a nudge the user can acknowledge and continue past, matching the California pack's 'offer break reminders' wording. The runtime's caps are enforcement: at the operator's configured limit, allow becomes enforce_break and the session pauses by rule.
Both gears exist because duties and products differ. The pack duty names reminders; many operators — schools, parents' expectations, their own judgement — want a firmer backstop behind the nudge. Caps supply it without pretending the law set the number. When you describe your setup publicly, keep the two gears distinct: promising enforcement and delivering reminders (or the reverse) is a protocol-page inaccuracy of the first order.
How the runtime tracks and fires
Caps are evaluated from the session lifecycle: the runtime sees session start, each message event with its timestamp, and session end, so it can measure elapsed session minutes and accumulate daily minutes per session reference (the one-way hash of the operator's session ID — the tracking key is pseudonymous by construction). On each decision the runtime compares elapsed values against the operator's caps for minor sessions; crossing either returns enforce_break, and the evidence event records the session and daily elapsed figures alongside the action.
Two scope rules matter. Caps bind only sessions the operator flags as minor — the flag is operator-asserted context, and the runtime does not guess ages. And caps are evaluated per the operator's configuration: change the configuration and you change behaviour from that moment, with the evidence chain preserving what the configuration produced before the change. The minor-mode guide covers asserting the flag; the integration guide covers the calls that feed the clock.
Choosing values you can defend
The product refuses to choose for you — deliberately, since a defensible cap is a function of your audience's age range, your partners' expectations and your session shape. The reasoning pattern that produces defensible values: start from the youngest user in scope, not the average; anchor the session cap well inside the 3-hour reminder cadence so enforcement and reminders tell one story (a reminder at 3 hours is incoherent if the session cap is 5); set the daily cap as a small multiple of the session cap, not an order of magnitude; and write the reasoning down where a reviewer can find it.
Whatever you choose, describe it as yours: 'operator-configured' is the product's own phrase, and it is the honest frame for the protocol page and for any regulator conversation. The packs require reminders and heightened safeguards; the caps are your implementation of care beyond the floor, and they read best presented exactly that way.
Reading enforce_break events afterwards
Enforcement events are operational data: their distribution tells you whether caps are calibrated or theatrical. enforce_break firing on a large share of minor sessions suggests caps set below your product's natural session length — friction that trains users to restart sessions rather than rest. Firing almost never across a large minor population suggests caps set as decoration. Both readings come from counts and elapsed values alone — the evidence carries the triggering figures precisely so calibration is a data conversation, not a vibes conversation.
Watch also for cap-shopping behaviour in aggregate: many short sessions per daily reference can indicate users restarting to dodge session caps — which the daily cap exists to bound. The pair of caps is a system; read them together. The minor-safety overview places caps among the full set of minor layers.
Five actions for caps that mean something
- Decide the age-knowledge basis for your minor flag before setting any cap — caps on a flag you cannot assert are theatre.
- Set the session cap inside the 3-hour reminder cadence so reminders and enforcement tell one coherent story.
- Set the daily cap as a considered multiple of the session cap, and write down the reasoning for both values in your ops records.
- Verify in staging: a minor session at the cap returns enforce_break, the evidence event carries the elapsed values, and an adult session is untouched.
- Review enforce_break distributions monthly and recalibrate deliberately — document changes and their dates like the policy decisions they are.
Worked example: calibrating caps for a school-hours companion
A fictional homework companion used in schools sets initial minor caps at 45 minutes per session and 2 hours per day — chosen for its youngest users (11-year-olds) and recorded as operator-configured in its partner documentation. First-month evidence: 6,900 minor sessions, enforce_break on 31% of sessions (nearly all at the session cap, clustered at 44–45 minutes), daily cap reached by 2% of daily references.
The team's reading: the session cap is doing real work but fighting the product's natural rhythm — students were mid-problem at the break. They raise the session cap to 60 minutes, keep the daily cap, and add a pre-cap reminder at 50 minutes using the snippet's reminder pattern, so enforcement lands as the end of a warned arc rather than a wall. Next month: enforce_break on 19% of sessions, completion rates up, daily-cap hits steady at 2%. The values are the team's own, the story invented; the mechanism, the evidence fields and the operator-configured framing are the product's.
Frequently asked questions
Who chooses the cap values?
The operator. Caps are operator-configured; the product presents no cap value as a statutory number. California's pack sets a break-reminder cadence (at least every 3 hours for known minors), not a session-length limit.
What happens when a cap is reached?
The runtime returns enforce_break instead of allow for that minor session, and seals an evidence event recording the action with the elapsed session and daily minutes that triggered it.
Do caps apply to all users?
No — only to sessions the operator flags as minor sessions. The minor flag is operator-asserted context; the runtime enforces caps against flagged sessions and does not guess ages.
How is usage tracked without identifying the minor?
Sessions are referenced by a one-way hash of the operator's session ID; elapsed session and daily minutes accumulate against that pseudonymous reference. No user ID, IP or message text is stored.
How do caps relate to break reminders?
Reminders (free kit) are offers — a gentle break suggestion at least every 3 hours. Caps (Pro/Team runtime) are enforcement — the session pauses at the operator's limit. Use both, and describe each accurately on your protocol page.
Keep going
Related guides: Minor Safety for AI Companions: Break Reminders, Caps and Parent Flags · Configuring Minor Mode: Asserting the Flag, Setting the Layers, Verifying the Behaviour · Crisis Evidence Runtime: Live Guardrail Decisions, Sealed as Evidence
Put it to work: Create the free kit · Eval harness · Scenario tests · Crisis runtime · Disclosure packs · Pricing · All CompanionGuard guides