> ## Documentation Index
> Fetch the complete documentation index at: https://hmm.heyhyper.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Token Usage & Costs — How Hyper Measures Memory Work
> Learn how Hyper tokens work, what each action costs, how to estimate your team's monthly usage, and what happens when you reach your plan limit.
Every interaction your agents have with Hyper's memory layer — recording a new fact or retrieving an existing one — draws from your workspace's monthly token budget. Tokens are a straightforward unit of measure: they reflect how much memory activity your team is generating, not how much compute power you're using. Understanding how tokens are counted makes it easy to size your plan correctly and avoid unexpected limits.
## What Is a Token?
A Hyper token represents a unit of memory work. When an agent stores something into Hyper's memory or pulls something back out, the operation costs a fixed number of tokens. Token balances reset at the start of each billing cycle, and unused tokens do not roll over.
Tokens are distinct from the LLM tokens used by your AI model provider (OpenAI, Anthropic, etc.). Hyper tokens measure memory operations only — your model's inference costs are billed separately by your model provider.
## Token Costs Per Operation
**2,000 tokens**
Triggered when an agent calls `remember` or `observe` to write a new fact into Hyper's memory. This includes any connector-driven ingestion that stores a new memory entry on your behalf.
**500 tokens**
Triggered when an agent calls `ask` to search and retrieve facts from Hyper's memory. Reads are significantly cheaper than writes, so your agents can query memory frequently without burning through your budget quickly.
## Plan Token Budgets
| Plan | Price | Tokens per Seat / Month |
| ---------- | -------------------- | ----------------------- |
| Free Trial | \$0 (3 days) | Unlimited |
| Sprint | \$30 / seat / month | 4,000,000 |
| Blitz | \$120 / seat / month | 20,000,000 |
| Enterprise | Custom | Custom / negotiated |
In a team workspace, each seat contributes its own token budget to the workspace pool. A five-person Sprint team has **20,000,000 tokens** available per month in total.
## Estimating Your Team's Usage
Use the breakdowns below to get a rough monthly estimate before choosing a plan.
**Step 1 — Count your write operations.**
Think about how many new facts your agents record per day. A moderately active agent might record 50 facts per day.
> 50 facts/day × 2,000 tokens × 30 days = **3,000,000 tokens/month** in writes
**Step 2 — Count your read operations.**
Reads are cheaper. An agent that retrieves facts 200 times per day uses far fewer tokens than its write footprint suggests.
> 200 retrievals/day × 500 tokens × 30 days = **3,000,000 tokens/month** in reads
**Step 3 — Multiply by your team size.**
If you have four agents running similar workloads, your total is roughly **24,000,000 tokens/month** — which fits comfortably in a four-seat Blitz plan (80,000,000 tokens combined).
When in doubt, start on Sprint. You can see your real-time consumption in **Settings → Billing** and upgrade to Blitz at any time mid-cycle.
## What Happens When You Hit the Cap
When your workspace exhausts its monthly token budget:
* **Write operations are blocked.** Calls to `remember` or `observe` will return an error until the next billing cycle begins or you upgrade your plan.
* **Read operations continue to work.** Calls to `ask` remain available so your agents can still retrieve existing memories even after the write cap is reached.
This means your agents never lose access to what Hyper already knows — they just can't record new facts until capacity is restored.
## How to Upgrade Your Plan
1. Open the Hyper desktop app.
2. Navigate to **Settings → Billing**.
3. Select the plan you want to upgrade to and confirm.
Upgrades take effect immediately. Your new token budget becomes available right away, and your billing date adjusts to reflect the prorated difference.
Compare Sprint, Blitz, and Enterprise side by side to find the right fit for your team's memory workload.