Understanding your token budget

Your subscription includes a monthly token budget — about 5 million tokens on Standard and 15 million on Pro, priced against MiniMax M2.7. That’s enough for the vast majority of users to do real daily work without thinking about it. But if you’re planning to run your agent hard — heavy scheduled tasks, big research runs, lots of sub-agents — your token budget is the throttle, and it’s important to understand how it works before your first end-of-month surprise.

This is the single most important page in Billing

If you skip everything else, read this page. It explains why a $29/month AI assistant can’t be literally unlimited — and what “unlimited-feeling” looks like in practice when you pay attention to your usage.

Why token-based

AI agents cost real money to run. Every time your agent thinks about something — reads an email, drafts a reply, researches a topic, browses a page, writes a plan — it sends tokens through a large language model, and those tokens cost the model provider real money. We pay them, and your subscription pays us.

We picked MiniMax M2.7 as the default model because it has one of the best price-to-capability ratios on the market — your token budget stretches much further than it would on Claude or GPT-4o at the same dollar value. That’s how we’re able to offer a real agent (not a chatbot) for $29/month.

What burns tokens

Activity	Typical token cost
A quick back-and-forth chat (“summarize this article”)	a few thousand tokens
A full inbox triage with drafts for 10 emails	tens of thousands of tokens
Reading a PDF and summarizing it	tens of thousands, depending on length
A scheduled morning briefing	~20-50k tokens per run
A deep research task (visits 15 pages, writes a 2-page briefing)	a few hundred thousand tokens
A 30-minute coding or debugging session with heavy tool use	1M+ tokens
Sub-agent research (spawns 5 helpers, each does its own browsing and analysis)	1-2M+ tokens per run

A few things to note:

• “Thinking about the page” is what costs — not the browsing itself. The sandbox browser and the Chrome extension are free. Load as many pages as you want. What costs is the LLM reading the page content and deciding what to do next.
• Scheduled tasks are sneaky. A morning briefing running daily at 40k tokens/run consumes ~1.2M tokens a month. Five scheduled tasks like that can eat your entire Standard budget before you realize.
• Sub-agents multiply fast. Each sub-agent uses its own token budget on top of the parent. One parent spawning 5 sub-agents for deep research can 6× your token cost per task.
• Tool use isn’t free inside the thinking step. Every time your agent decides to run a tool (send email, read file, navigate browser), that decision is made by an LLM call. Long tool-use chains add up.

What does NOT burn tokens

• Web browsing itself. Both the sandbox browser and the Chrome extension are free. Zero token cost for the page load, the click, the type.
• Reading your memory files (MEMORY.md, USER.md).
• Reading your workspace files.
• IMAP/CalDAV sync (your email and calendar providers are free to us).
• Telegram / Discord / Slack message delivery.
• UI interactions — clicking around the dashboard.
• Credit pack purchases and other billing activity.

Real-world usage patterns

Here’s what actual usage looks like across different user profiles, based on real patterns:

The “light user” — Standard plan, fits comfortably

• Daily morning briefing (30k tokens/run × 30 days = ~900k)
• ~10 real conversations per week averaging 20k tokens each (~900k)
• Weekly competitor monitor (50k/week × 4 = ~200k)
• One-off tasks, occasional research, email drafting (~1M)
• Monthly total: ~3M tokens ✅ Well under the 5M Standard budget

The “power user” — Standard plan, starting to pinch

• Morning + weekly briefings (~1.5M)
• ~30 real conversations per week at 25k each (~3M)
• 3 daily competitor monitors (~600k)
• Weekly deep research with one sub-agent (~400k)
• Monthly total: ~5.5M tokens ⚠️ Over Standard, fine on Pro, or top up

The “heavy user” — Pro plan territory

• Multiple daily briefings and full inbox triages (~3M)
• 50+ conversations per week at 30k each (~6M)
• Daily deep research with sub-agents (~3M)
• A handful of monitors, reports, content workflows (~2M)
• Monthly total: ~14M tokens ✅ Pro’s 15M budget fits

The “running it like a company” — Pro + BYOK or top-ups

• Continuous multi-agent workflows, 24/7 monitors, daily research at scale
• 30M+ tokens/month
• Solution: Either buy credit packs to extend, or bring your own API key (BYOK). On BYOK, your agent routes requests directly through your own Anthropic/OpenAI/MiniMax key and doesn’t touch our metered budget at all.

How to stay in budget

Watch the credit-alert banner

Your agent warns you at 80% consumption and again at 100% via your notification channels. If you see the warning early in the month, you’re burning faster than your plan covers — time to adjust.

Check your usage page

Dashboard → Usage shows your current month’s consumption, a breakdown by source (chat vs. cron vs. sub-agents vs. search), and your projected end-of-month burn. A 30-second weekly check is enough to stay ahead of surprises.

Review your scheduled tasks

Most surprise bills come from a cron job that runs more often or does more work than the user meant. Go to Dashboard → Scheduler, look at the “tokens used” column for each task, and pause anything that’s consuming more than you expected. You can resume later.

Be specific about sub-agents

“Research this for me” costs less than “research this in depth with 5 parallel sub-agents.” The latter is sometimes what you want — deep research runs, big monitoring sweeps — but use it deliberately, not by accident.

Cache your memory

Every time your agent starts a new conversation, it reads its memory files. Dense memory files = more tokens per conversation. If you see MEMORY.md growing to 10k+ lines, ask your agent to compact it.

When you run out

Your agent does not suddenly stop working when you hit 100%. The behavior is gradual:

1. At 80%: Warning notification. Your agent is still fully operational.
2. At 100%: Exhaustion notification. New heavy work (deep research, sub-agents, big scheduled tasks) is throttled. Direct conversations and simple questions keep working.
3. Until next cycle: You can continue using your agent for basic work. Top up with a credit pack anytime to restore full capacity.

Three paths when you need more

1. Credit packs — Buy additional tokens at any time from Settings → Billing → Credit packs. Never expire, stack on top of your monthly allowance. Good for occasional bursts.
2. Upgrade to Pro — 3× the included tokens ($5M → $15M) plus higher limits across the board. Better if the bursts become the norm.
3. BYOK (bring your own API key) — Route your agent through your own Anthropic, OpenAI, MiniMax, DeepSeek, or OpenRouter key. Your agent uses the provider directly and doesn’t touch your included allowance at all. You pay the provider’s bill, not ours. Available on every tier including trial.

The flat-rate path for heavy users

If your usage is consistently above the included monthly credit — say you’re a real estate agent answering 500+ Zillow leads, an ops team running multiple daily briefings, or anyone with constant scheduled research — the cleanest economics are:

1. Subscribe to MiniMax’s flat-rate coding plan (~$59/month at the time of writing — check their site for current rate-limit windows).
2. Drop the API key into Settings → Billing → BYOK → MiniMax.
3. Your agent now runs on the flat-rate plan. Within MiniMax’s per-window rate limits, your inference is effectively unlimited. Our included credits sit untouched as a backup.

Total monthly cost for an unlimited-feeling assistant: your tier (e.g. $29 Standard) + MiniMax’s flat-rate fee. No metering, no surprise bills. This is what the platform’s own founder runs personally.

The honest pitch

Here’s what we want you to understand going in:

• Our agent is extremely capable for the price — capable enough that you can absolutely burn through a month’s tokens in a week if you run it like an enterprise tool.
• $29/month is not “unlimited AI employee.” It’s “a very generous working budget for an AI employee that handles most real daily work for most real users.” Those are different claims and we want to be clear about which one we make.
• We’d rather you understand this and stay happy than discover it after your first throttle. If you know what burns tokens and you watch the usage page, you’ll almost never hit the wall. If you don’t, and you set up 10 deep-research cron jobs on day one, you will.
• The tools to go bigger exist. Credit packs, Pro tier, BYOK. All three are available on day one. None of them are hidden behind a “talk to sales” wall.

If anything in this page is confusing or you hit an unexpected throttle, email support. We’d rather help you right-size than lose you to an avoidable surprise.