# Shardflux

Official pricing: https://shardflux.dev/#pricing  
Category: agent-sandbox · Isolation: firecracker

## Pricing regimes (raw)

- **Workspace** (resource): $0.12/vCPU-h, $0.04/GiB-h
- **File-first workspace (fresh VM per command)** (resource): $0.12/vCPU-h, $0.04/GiB-h [alt]

## 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), root, browser, browser automation API, egress allowlist, open internet, Python, Node.js, wake on request, live fork (no pause), memory fork, audit logs, spend limits, MCP server, automatic snapshots, snapshots on demand, gVisor or VM (no shared kernel), firewall inside (nftables), extra volumes, shared volumes

No: nested virtualization, desktop GUI, computer-use API, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, preinstalled agents, SSH, public IPv4, HTTPS preview URLs, static egress IP, self-hosting, BYOC, open source, GPU, arm64, Windows, macOS, HTTP method/path egress rules, live resize, EU data residency, agent harness API, hosted agent API (their own agent), inbound access rules, secret proxy, secret proxy for any API

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

## Caveats

- Plans include CPU-hours and RAM GiB-hours, not dollars. They are valued at the published overage rates ($0.12/CPU-hour, $0.04/GiB-hour) so the engine can net them against usage; in reality the two allowances do not offset each other, so a workload that uses up only one of them can cost somewhat more than shown.
- Overage is off until you turn it on, with a spend cap you set (shardflux.dev pricing, 2026-09-29).
- Retained storage (suspended workspaces, shared volumes) is never charged; outgoing transfer has no overage price and is blocked when the plan allowance runs out, so egress beyond the allowance is priced as unknown.
- Every workspace is its own microVM with its own kernel; the hypervisor is not named. Suspend checkpoints memory and processes; resume p50 0.3 s, new workspace p50 0.2 s (vendor claims on the landing page).
- Templates (e.g. python-node-browser) are the base images; whether you can bring a Dockerfile/OCI image is not documented. Docker inside, nested virtualisation, root and SSH are not documented.
- Workspaces run in AWS us-west-2 only; no data residency outside the United States. Start waits up to 15 minutes for capacity before failing with capacity_unavailable.
- Features set to null were not verified. No workspace was purchased or tested; feature/performance statements are documentation claims.
- Opt-in overage is capped at most at the plan price ($9/$79/$299), so a month's bill is at most 2x the plan fee; beyond that, starts are refused and running workspaces are suspended (402 spend_cap_reached). Large workloads need Enterprise.
- Idle-but-running workspaces bill full RAM: hosts 'park' idle workspaces but they stay running and use their full allocation. The adaptive idle timeout is 5 min until learned (10 s to 4 h), or it can be set to fixed:60-604800 s or never.
- Rate limits: 3,600 tool tokens per hour per API key (an open that returns a token counts; 429 rate_limited beyond), 3 concurrent template builds per org.