# Kedge

Official pricing: https://kedge.dev/docs/billing  
Category: agent-sandbox · Isolation: vm

## Pricing regimes (raw)

- **Persistent machine/workspace (preview price estimates)** (resource): $0.02054794520547945/vCPU-h, $0.00684931506849315/GiB-h [beta, alt]
- **Disposable and held sandbox API (public preview)** (resource): $0.02054794520547945/vCPU-h, $0.00684931506849315/GiB-h [beta]

## Features

Yes: snapshots, memory snapshots, fork/clone, pause/resume, persistent disk, volumes, idle auto-stop, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, your own Docker/OCI image, full VM (own kernel), Docker inside, preinstalled agents, SSH, HTTPS preview URLs, custom domains, egress allowlist, open internet, private networking, Python, Node.js, credential injection, wake on request, live fork (no pause), memory fork, webhooks, scheduled runs (cron), EU data residency, scoped API keys, spend limits, automatic snapshots, snapshots on demand, agent harness API, gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), extra volumes, shared volumes

No: nested virtualization, browser, desktop GUI, computer-use API, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, public IPv4, raw TCP inbound, static egress IP, self-hosting, BYOC, open source, SOC 2, HIPAA, SSO, GPU, arm64, Windows, macOS, HTTP method/path egress rules, live resize, MCP server, hosted agent API (their own agent), secret proxy, secret proxy for any API

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

## Caveats

- Rates are estimates; actual paid billing had not started as of the Show HN (2026-07-29) and no evidence of paid top-ups since.
- Maximum cpus / memory per VM and per-account concurrency not published.
- Hypervisor not named ('hardware-isolated VM with its own Linux kernel').
- Whether deploy snapshots / warm-pool clones count toward billed storage; whether memory billing of CoW-shared pages is per clone.
- Ingress pricing not stated; public IPv 4 / static addresses not offered.
- Coding-agent model token rates not published.
- This mode is persistent machines/workspaces, not the disposable sandbox API capped at 1 hour. Flags beta+alt prevent default selection.
- Memory snapshot billing remains unknown; retained volumes use 0.05/GiB-month separately.
- An account can hold only a limited, unpublished number of sandboxes at once. Past the cap, create fails with a limit error rather than queuing.
- Volumes publish local writes to object storage after 1 s or 16 MiB. The local writeback buffer is 4 GiB, and new writes fail when it fills.

## How this provider charges


As of 2026-09-28. [Public preview launch July 29 2026](https://news.ycombinator.com/item?id=49099434). [Billing docs](https://kedge.dev/docs/billing) publish estimates; launch author said paid billing was not live. No paid invoice validated.

| Regime | Applies when | Meter / numbers | Source |
|---|---|---|---|
| Active resource use | Running VM | CPU 15 USD/vCPU-month ÷730 = 0.0205479452/vCPU-h; RAM 5 USD/GiB-month ÷730 = 0.0068493151/GiB-h; actual CPU and working-set RAM | [Billing](https://kedge.dev/docs/billing) |
| Suspended / scaled-zero | Persistent machine or app idle | No compute; stored bytes continue | [Runtime](https://kedge.dev/docs/runtime) |
| Storage | Retained written data |0.05 USD/GiB-month; VM memory snapshot price unknown | [Billing](https://kedge.dev/docs/billing), [volumes](https://kedge.dev/docs/volumes) |
| Egress | Outbound |0.01 USD/GiB | [Billing](https://kedge.dev/docs/billing) |
| Preview allowance | Every account |5 USD/month resource credit; separate 2 USD agent-model allowance | [Billing](https://kedge.dev/docs/billing) |
| Disposable sandbox | One-shot or held execution |5-minute idle expiry; held TTL at most 1 hour;10-minute command limit | [Sandboxes](https://kedge.dev/docs/sandboxes) |

## Gotchas

Rates are preview estimates, not timeless list prices. Hypervisor unnamed. Sizes and concurrency ceilings unpublished. CPU caps are not reservations; RAM uses working set. App deploy snapshots do not establish billable snapshot storage at the volume price. Engine card intentionally models persistent machines/workspaces under alt+beta flags; it does not imply 8-hour disposable sandboxes.

## Required worked example

50 ×8 h/day ×22 days =8,800 hours,4 vCPU/8 GiB,30% CPU. Disposable sandbox cannot remain held for 8 hours. For the **alternative persistent workspace**, assuming 8 GiB billable working set and sufficient capacity, compute is `8800 × (4×.30×15/730 +8×5/730)` = **$699.178082**.100 GiB egress adds$1. Retained 50 GiB ordinary volume would add$2.50, but is not a price for 50 GiB memory snapshots.5 USD monthly resource credit may be deducted once if applicable; models/unknown snapshot cost remain outside. No complete confirmed bill.