Plans, Quotas and Usage¶
Every account is on a plan. The plan sets hard limits, and AgntSpark meters what you use.
Alpha
These are the limits during the invite-only alpha. They, and the plans themselves, may change before general availability.
Limits¶
| Free | Pro | |
|---|---|---|
| Price | $0 | $29 per month |
| Agents | 3 | 20 |
| Running replicas, all agents | 3 | 20 |
| vCPU across running replicas | 3 | 16 |
| Memory across running replicas | 1.5 GB | 8 GB |
| vCPU per replica | 1 | 2 |
| Memory per replica | 1 GB | 4 GB |
| Requests per minute per agent URL | 300 | 3,000 |
Prices are in US dollars, billed monthly, plus any applicable tax. You bring your own model provider key, so model usage is billed by your provider, not by AgntSpark.
CPU limits cap how much CPU your replicas may use; memory limits are hard, and a replica that exceeds its memory is restarted.
Your current plan, limits and usage are on the console dashboard and at
GET /v1/account/usage.
How quotas are enforced¶
Resource quotas are checked before anything happens. Creating,
deploying or scaling an agent past a limit returns 403 QUOTA_EXCEEDED, with
details naming the resource, your limit, what's current and what was
requested, and nothing changes. Auto-scaling stops at your limits.
Running agents keep running if you change to a plan with lower limits; you just can't add more until you're back under them.
The per-agent request limit is enforced per minute at the agent's URL; see Access Keys and Rate Limits.
What's metered¶
Usage is recorded per agent, per hour:
| Measure | How it's counted |
|---|---|
| Replica-hours | Running replicas × time running |
| vCPU-hours | Running replicas × vCPU requested per replica × time |
| Memory GB-hours | Running replicas × memory requested per replica × time |
| Requests | Requests the platform passed to the agent |
Resources are metered on what you reserve, the cpu and memory_mb in
your deploy config, not what your process happens to use. Stopped agents use
nothing.
Requests refused by the platform aren't counted: no valid access key (401),
rate limited (429), or no running replica (503). Everything else is,
including automated traffic such as scanners, which find public agent URLs
quickly. A private agent only counts requests that carry one of its keys.
Changing plans¶
Upgrade under Settings → Plan & Billing; payment is handled by Stripe. Manage your card, invoices and cancellation from the same page. A cancellation takes effect at the end of the billing period, after which the account returns to the free plan's limits.