# Koyeb

Sandboxes started from a Python or JavaScript SDK on Koyeb's microVM instances, billed per second at the instance-type price.

- Operated by: Koyeb, Inc.
- Status: Active [1]
- Website: https://www.koyeb.com
- Categories: Agent sandboxes
- Checked: 31 Oct 2026

## Koyeb Sandboxes (Agent sandboxes)

- Lowest price: $0.0144 /h (small (1 vCPU / 1 GB))
- 1-hour session: ≥ $0.0288 (2 vCPU / 4 GB · rank 4 of 26)
- Fleet month: ≥ $253.44 (50 × 8 h × 22 days)
- Billed on: Running time (per second)
- Isolation: unknown (max session unknown)

### Plans

- Starter: Free; Pay-as-you-go; sandboxes available [2]
- Pro: $29 /month; $10 of compute a month, 10 users, 100 services, 1 TB bandwidth [6]
- Scale: $299 /month; $100 of compute a month, 50 users, 1,000 services, AWS regions, 99.9% SLA [6]

### Billing

- Payment: Metered: usage billed afterwards [2]
- Plan minimum: None [2]
- Billing increment: Per second [2]
- Minimum charge: unknown (Not published) [2]
- Per-job fee: None [2]

### Runtime

- Billed on: Running time [2]
- Isolation: unknown (Instances are microVMs; the hypervisor is not named) [2]
- Max session: unknown (No lifetime limit is published; auto-delete delays can be set up to 24 hours) [2]
- Concurrent sandboxes: unknown (Not published) [2]
- Largest vCPU: 40 [2]
- Largest memory: 128 GB [2]

### Lifecycle

- Survives a stop: Nothing (ephemeral) [4]
- Snapshots: unknown (Not documented for sandboxes) [4]
- Pause and resume: Yes [4]
- Fork: unknown (Not documented for sandboxes) [4]
- Compute billed while stopped: unknown (idle_timeout sends a sandbox to deep sleep; billing while asleep is not stated) [4]
- Auto-stop: unknown (idle_timeout is configurable; the default is not stated in the docs) [4]
- Start time (vendor claim): unknown (No sandbox start time published; the platform claims cold starts as low as 250 ms) [4]

### Capabilities

- GPU: Yes [5]
- Desktop: None [5]
- Coding agents installed: Mistral Vibe, Codex, Gemini CLI, OpenCode [5]
- Agent prompt API: No [5]
- SSH: unknown (Not documented for sandboxes) [5]
- Docker inside: unknown (Not documented for sandboxes) [5]
- SDKs: Python, TypeScript [5]
- SOC 2: Yes [5]

### Storage and network

- Disk included: unknown (Included per instance type (2 GB to 400 GB SSD); see sizes) [6]
- Extra disk: unknown (No volume price for sandboxes is published) [6]
- Snapshot storage: unknown (Not published) [6]
- Egress: $0.02 per GB [6]

### Platform

- Regions: fra, was, sin, tyo, par, aws-us-east-1, sfo [3]
- Your cloud (BYOC): unknown (Not stated for sandboxes) [3]
- Self-host: unknown (Not stated) [3]

### Free allowance

- Minutes a month: None [2]
- Credit a month: None [2]
- One-time credit: None [2]

### Prices

| Size | vCPU | RAM | Disk | Plan | Price | Per hour |
|---|---|---|---|---|---|---|
| small | 1 | 1 GB | 10 GB | any | $0.000004 /s [7] | $0.0144 |
| medium | 2 | 2 GB | 20 GB | any | $0.000008 /s [7] | $0.0288 |
| large | 4 | 4 GB | 40 GB | any | $0.000016 /s [7] | $0.0576 |
| xlarge | 8 | 8 GB | 80 GB | any | $0.000032 /s [7] | $0.115 |
| 2xlarge | 16 | 16 GB | 160 GB | any | $0.000064 /s [7] | $0.23 |
| 3xlarge | 24 | 32 GB | 240 GB | any | $0.000128 /s [7] | $0.461 |
| 4xlarge | 32 | 64 GB | 320 GB | any | $0.000256 /s [7] | $0.922 |
| 5xlarge | 40 | 128 GB | 400 GB | any | $0.000512 /s [7] | $1.84 |
| eco-medium | 1 | 2 GB | 16 GB | any | $0.000004 /s [7] | $0.0144 |
| eco-large | 2 | 4 GB | 20 GB | any | $0.000008 /s [7] | $0.0288 |
| eco-xlarge | 4 | 8 GB | 20 GB | any | $0.000016 /s [7] | $0.0576 |
| eco-2xlarge | 8 | 16 GB | 20 GB | any | $0.000032 /s [7] | $0.115 |
| RTX-4000-SFF-ADA (NVIDIA RTX 4000 SFF Ada) | 6 | 44 GB | 240 GB | any | $0.5 /h [6] | $0.5 |
| L4 (NVIDIA L4) | 6 | 32 GB | 160 GB | any | $0.7 /h [6] | $0.7 |
| RTX-A6000 (NVIDIA RTX A6000) | 6 | 64 GB | 160 GB | any | $0.75 /h [6] | $0.75 |
| L40S (NVIDIA L40S) | 15 | 64 GB | 160 GB | any | $1.20 /h [6] | $1.20 |
| A100 (NVIDIA A100 80GB) | 15 | 180 GB | 320 GB | any | $1.60 /h [6] | $1.60 |
| A100 SXM (NVIDIA A100 SXM 80GB) | 15 | 180 GB | 320 GB | any | $2.15 /h [6] | $2.15 |
| RTX-PRO-6000 (NVIDIA RTX PRO 6000) | 15 | 220 GB | 240 GB | any | $2.20 /h [6] | $2.20 |
| H100 (NVIDIA H100) | 15 | 180 GB | 320 GB | any | $2.50 /h [6] | $2.50 |
| H200 (NVIDIA H200) | 15 | 240 GB | 320 GB | any | $3.00 /h [6] | $3.00 |
| B200 (NVIDIA B200) | 15 | 240 GB | 320 GB | any | $5.50 /h [6] | $5.50 |

- Sandboxes are in public preview and available on the Starter, Pro and Scale plans. They run on the same instance types and per-second prices as other Koyeb services.
- The default instance type for a sandbox is micro (0.5 vCPU, 512 MB, $0.000002/s). Sizes with fractional vCPU (nano, micro, eco-nano, eco-micro, eco-small) are not listed here.
- Eco instance types are offered in Washington, D.C., Frankfurt and Singapore only.
- Bandwidth above the included 1 TB costs $0.02/GB in EU/US and $0.04/GB in Asia (pricing page); the pricing FAQ says outbound bandwidth is not yet charged and quotes $0.04/GB once billing begins.
- GPU use in sandboxes is stated in the launch post; GPU prices come from the pricing page (per hour). Multi-GPU instance types (2x, 4x, 8x) are not listed here.
- SOC 2 and ISO 27001 are listed under the Enterprise plan on the pricing page.
- The default image includes Mistral Vibe, Codex, Gemini CLI and OpenCode.

## Sources

[1] https://www.koyeb.com/docs/sandboxes (checked 31 Oct 2026)
[2] https://www.koyeb.com/docs/faqs/pricing (checked 31 Oct 2026)
[3] https://www.koyeb.com/docs/reference/regions (checked 31 Oct 2026)
[4] https://www.koyeb.com/docs/sandboxes/sandbox-lifecycle (checked 31 Oct 2026)
[5] https://www.koyeb.com/docs/sandboxes/sandbox-default-image (checked 31 Oct 2026)
[6] https://www.koyeb.com/pricing (checked 31 Oct 2026)
[7] https://www.koyeb.com/docs/reference/instances (checked 31 Oct 2026)