# Amp Orbs

Official pricing: https://ampcode.com/docs/orbs/sizes-and-costs  
Category: dev-env · Isolation: vm

## Pricing regimes (raw)

- **Hosted orb compute, PAYG or subscription allowance** (sizes): a1.tiny 1 vCPU/2 GiB $0.08/h; a1.small 2 vCPU/4 GiB $0.17/h; a1.medium 4 vCPU/8 GiB $0.33/h; a1.large 8 vCPU/16 GiB $0.66/h; a1.xxlarge 16 vCPU/32 GiB $1.32/h; a1.3xlarge 16 vCPU/44 GiB $2.13/h [alt]

## Features

Yes: snapshots, memory snapshots, pause/resume, persistent disk, idle auto-stop, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, full VM (own kernel), Docker inside, root, systemd, browser, desktop GUI, computer-use API, browser + desktop control, preinstalled agents, HTTPS preview URLs, custom domains, self-hosting, SOC 2, SSO, Python, Node.js, wake on request, scheduled runs (cron), audit logs, spend limits, automatic snapshots, agent harness API, gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables)

No: fork/clone, volumes, your own Docker/OCI image, nested virtualization, anti-bot stealth, CAPTCHA solving, residential IPs, SSH, public IPv4, egress allowlist, static egress IP, BYOC, open source, HIPAA, GPU, arm64, Windows, macOS, credential injection, HTTP method/path egress rules, live resize, live fork (no pause), memory fork, EU data residency, MCP server, snapshots on demand, hosted agent API (their own agent), secret proxy, secret proxy for any API, extra volumes, shared volumes

Unknown: everything else. Evidence (source + quote) per feature: https://battleships.dev/data/providers/amp.json → feature_evidence

## Caveats

- Underlying isolation technology and CPU sharing
- Concurrent running-orb quota (spawn pacing is not concurrency)
- Exact cross-size subscription allowance conversion and overage bucket mechanics
- Disk overage and transfer overage; repository storage is not orb disk
- No guarantee 50 orbs can start simultaneously; default burst is 20.
- Unknown storage, egress, min charge and start fee must not be silently treated as confirmed zero.
- Subscription plans are size-restricted representations of one real subscriber tier; choose only one, never stack variants. included_usd covers compute only, not storage/egress/models.
- Per-user orb allowances do not pool. Card accurate for one subscriber and uniform orb size; multi-seat or mixed-size workloads need explicit per-user/minute accounting.
- Model allowance ($20/$200) is separate and not subtracted from compute.
- a1.3xlarge requires Gigawatt/Enterprise; Hobby/Megawatt plans exclude it.
- Default E2B microVM orbs documented in GCP us-west1; no specific hypervisor assumed. Self-hosted runner support does not change hosted price.
- Enterprise orb rates may differ from list: 'Orb pricing for enterprise workspaces is shown in project settings.'
- Idle tail: orb bills until 5 min after agent work ends or 20 min after last user interaction (whichever later). Opening an orb thread in web/app, any authorized portal HTTP request (incl. from a service/uptime check) wakes the orb and starts a new 20-min billed window; Orb Saver Mode stops thread-opening from waking it. Archiving a thread pauses immediately. Startup time is billed.
- Education discount: students/teachers get Megawatt for $10/month (same 45,000 small-orb minutes, $10 of included credits); Megawatt only, eligibility-gated so not modelled as a general plan.
- Subscription allowance is one per-user bucket across sizes (larger orbs consume it faster), per-person and not pooled in workspaces; it resets monthly, does not roll over; beyond it orbs are paid from credits at standard price. Subscriptions are for a single User. Teams workspace has no fee but SSO/custom domains need at least one paid member.

## How this provider charges


As of 2026-09-28. Category: `agent-platform`. USD list prices unless explicitly labelled otherwise.

| Regime | When / billing unit | List price | Source |
|---|---|---|---|
| Orb PAYG | Running orb, metered per minute | CPU/RAM: 1/2 $0.08/h; 2/4 $0.17; 4/8 $0.33; 8/16 $0.66; 16/32 $1.32; 16/44 $2.13; each has 60 GB disk | [Sizes](https://ampcode.com/docs/orbs/sizes-and-costs) |
| Megawatt | $20/month | 750 small (2/4) orb-hours, plus at least $20 agent usage; 1 GB repository storage, 10 GB monthly transfer | [Pricing](https://ampcode.com/docs/pricing) |
| Gigawatt | $200/month | 1000 xxlarge (16/32) orb-hours, plus at least $200 agent usage; 10 GB repository storage, 100 GB transfer | [Pricing](https://ampcode.com/docs/pricing) |
| Unconstrained / Enterprise | Credits without subscription / contract | PAYG models at API prices for non-enterprise, separate orb usage; Enterprise custom | [Pricing](https://ampcode.com/docs/pricing) |

## Caps and gotchas
`a1.3xlarge` needs Gigawatt or Enterprise. Paused orbs cost nothing; files/services persist. Pause requires both 5 minutes without agent work and 20 minutes without user/portal interaction. Opening a portal wakes the orb; background dev server alone doesn't prevent pause. Minute billing includes startup. [Sizes/lifecycle](https://ampcode.com/docs/orbs/sizes-and-costs)
Twenty starts may burst per user, then one per five minutes; additional starts queue. That's not a 20-running-orb limit. [Orbs](https://ampcode.com/docs/orbs)
Separate **orb allowance** and **agent/model allowance**: never subtract $200 model usage from compute. Expanded pricing-page tables specify exact minutes by size: **medium gets 22,500 minutes (375h) on Megawatt or 240,000 minutes (4000h) on Gigawatt**. Allowances are per person, not pooled across team members. Included usage does not roll over; purchased credits expire after 12 months. Retained external snapshot/export and disk/egress overage rates unknown. [Usage](https://ampcode.com/docs/pricing)

## Worked examples
PAYG 4/8 for 8,800 running hours: **$2,904**, plus models/tools, startup/idle rounding and unknown extras. CPU 30% doesn't change this allocation price. Fifty simultaneous new starts exceed the default burst; negotiate/queue rather than assume immediate capacity. Single developer 176 hours on medium: **$58.08 PAYG compute** plus models.
Subscriptions can be cheaper. The expanded official pricing table gives Gigawatt **4000 medium-hours**: $200 + (8800−4000)×$0.33 = **$1,784 compute plus subscription**, for one subscriber owning this entire homogeneous workload, before models/extras and quota confirmation. Megawatt gives 375 medium-hours: $20+(8800−375)×$0.33 = $2,800.25. For 176 medium-hours, Megawatt's $20 covers compute within allowance, plus any model spend beyond its separate allowance.

The engine card encodes per-size subscription minute grants as compute-only dollar offsets restricted to that size. These are alternate representations of one tier, not stackable plans. Multi-person workloads need per-user accounting because minutes are not pooled. Model dollars must never offset compute a second time. See `raw/amp-gigawatt-expanded.txt` and the expanded Megawatt table in `raw/browser/ampcode_com_pricing.txt`.

Default hosted orbs are documented as E2B microVMs in GCP us-west1; setup documentation uses Debian 12 and sudo. Exact CPU sharing/hypervisor details are not inferred. [Customization](https://ampcode.com/docs/orbs/customizing)

For **three real Gigawatt members**, each owning a balanced 2,933.33-hour medium workload, all 8,800 hours can fit their separate 4,000-hour allowances for $600/month base. This is a changed multi-user workload, not pooled credit or permission to share a personal subscription; per-user pacing still applies. Model credits remain separate.