Skip to main content

Why caps exist

AI agents have an uncomfortable talent: spending very fast. An ambiguous prompt, an unexpected loop, an integration that keeps returning errors while the agent retries — any of these scenarios can burn through Stars in minutes. Caps are hard limits you configure before any spending happens, ensuring that even in the worst case your usage stays under control.

The four protection layers

Apollo Space applies caps in layers. Before each billable call, the system checks:

1. Wallet balance

The fundamental safeguard — if the wallet is empty, no new calls are authorized. You can operate freely as long as the balance is positive.

2. Monthly cap per organization

Even with a positive balance, you can set a monthly spending limit for the entire org. Once reached, all paid calls stop until the next cycle (or until you raise the cap).

3. Daily cap per agent

Limits how much each agent can spend per day. Useful for preventing a single looping agent from consuming the entire day’s budget.

4. Monthly cap per user × agent

The most granular combination — how much user X using agent Y can spend in a month. Useful on larger teams where you want to allocate budget per person.

How to configure

Everything lives in Settings → Spending Caps inside the app.

Org cap

Configured once by the org owner/admin. Typically revisited every few months as actual usage evolves.

Agent cap

Configured per agent when creating or editing a persona. Sensible defaults are applied automatically (you don’t need to start from scratch); customize only when the default doesn’t fit.

User × agent cap

Configured optionally by the admin. You don’t need to fill it in for every pair — only for specific users you want to limit explicitly (e.g., an intern who needs a boundary, a premium agent that only authorized people should use).

Precedence

All 4 layers are checked in order. The first protection that fires blocks the call — the rest are not even evaluated. Example: if the balance is empty, it doesn’t matter whether the org cap still has headroom or the user still has quota — the call is blocked at layer 1. This means you don’t need to configure all 4 layers. Layer 1 (balance) already protects you from the worst case; the others are fine-tuning tools for specific scenarios.

What the user sees when a cap fires

When a call is blocked by a cap, the agent:
  • Stops execution immediately (no retry)
  • Surfaces a clear message in the chat indicating which cap was reached (e.g., “This agent’s daily limit has been reached — you can raise it in Settings → Caps”)
  • Suggests the corrective action (raise the cap, wait for the next cycle, switch to another agent)
No paid call ever happens “silently” — users always know when they’ve hit a limit.

Audit

Every time a cap fires, it is recorded in the org’s event history with:
  • Which cap triggered (balance / org / agent / user×agent)
  • What the configured value was
  • How much the call would have cost
  • Who triggered it (which user + which agent)
  • Timestamp
Useful for spotting patterns — if a cap is firing frequently, the configuration may need adjustment. For teams just getting started:
1

Trust the balance + initial bonus

Don’t configure anything yet. The initial bonus is generous enough for comfortable exploration.
2

Once you go into real operations, add agent caps

Set a daily limit for your most-used agents (Marcus, Athena). A reasonable default — you can adjust later as actual usage becomes clear.
3

Once you pass 5 active users, add an org cap

A monthly limit for the entire org prevents surprises at the end of the month. Calculate it based on 70–80% of your plan’s monthly credit.
4

User × agent caps only for specific cases

Don’t default this for everyone. Use it only when governance requires it (e.g., an intern, a premium agent).

Next steps

Stars — overview

How the wallet that caps protect actually works.

Plans and packs

How to size your balance and buy additional credits.

Agents

Every agent respects caps automatically — no additional code required.