# MIOSA

Official pricing: https://miosa.ai/pricing  
Category: agent-sandbox · Isolation: firecracker

## Pricing regimes (raw)

- **Sandbox (fixed sizes xs-xl, managed)** (sizes): xs 1 vCPU/2 GiB $0.080667/h; small 2 vCPU/4 GiB $0.155187/h; medium 4 vCPU/8 GiB $0.310374/h; large 8 vCPU/16 GiB $0.620748/h; xl 16 vCPU/32 GiB $1.241496/h
- **Computer (durable Xfce desktop, fixed sizes)** (sizes): xs 1 vCPU/2 GiB $0.080667/h; small 2 vCPU/4 GiB $0.161334/h; medium 4 vCPU/8 GiB $0.328815/h; large 8 vCPU/16 GiB $0.65763/h; xl 16 vCPU/32 GiB $1.31526/h
- **Bring your own cloud (45% orchestration fee only, + your cloud bill)** (sizes): xs 1 vCPU/2 GiB $0.0363/h; small 2 vCPU/4 GiB $0.069834/h; medium 4 vCPU/8 GiB $0.139668/h; large 8 vCPU/16 GiB $0.279337/h; xl 16 vCPU/32 GiB $0.558673/h [alt, sales, 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), root, systemd, browser, desktop GUI, computer-use API, browser + desktop control, preinstalled agents, SSH, HTTPS preview URLs, custom domains, egress allowlist, open internet, BYOC, Python, Node.js, credential injection, HTTP method/path egress rules, live fork (no pause), memory fork, webhooks, audit logs, MCP server, automatic snapshots, snapshots on demand, agent harness API, hosted agent API (their own agent), fast boot (benchmarked), gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), secret proxy, secret proxy for any API, extra volumes

No: Docker inside, nested virtualization, code interpreter, anti-bot stealth, CAPTCHA solving, residential IPs, public IPv4, static egress IP, self-hosting, open source, SOC 2, HIPAA, SSO, GPU, arm64, Windows, macOS, GPU desktop, wake on request, live resize, EU data residency, shared volumes

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

## Caveats

- Rates are exact credit prices (1 credit = $0.01): vCPU 4.536, RAM 1.458/GB-h, disk running 0.06147/GB-h, cold storage 0.00243/GB-h, egress 5/GB. Page rounds vCPU to $0.0454 and RAM to $0.0146.
- Disk-while-running is folded into each size hour; while paused/stopped the disk bills at cold-storage $0.01774/GB-month (encoded as snapshot_gib_month). Whether a paused memory image adds to the GB counted is unpublished.
- Per-computer plan caps (Dev 2/4/10, Business 8/16/40, Enterprise 16/32/80) were on the pricing page earlier on 2026-09-28 and are absent on re-check; kept in plans as they may still be enforced.
- Stale conflicting docs: /docs/api-reference/credits still lists an old model (Small 4GB/1CPU 3 credits/h, Medium 6, Large 12; plans Free/Starter $29/Pro $79/Scale $199; credits expire 180 days). Ignored in favour of the pricing page.
- Plan fee = monthly minimum credited 1:1, so Business/Enterprise cost max(fee, usage).
- Docs: default 10 concurrent sandboxes per tenant (raise via support) vs. plan pages 2/50/200 concurrent workloads.
- Billing granularity/minimum unpublished (metered continuously, ledger settles ~5 min). GB vs GiB not specified (page says GB).
- BYOC mode excludes the underlying cloud bill; gated private preview. GPU rate $1.75/h is on the card but managed GPU is not shipped.
- Regions: 3 US regions (us-west-la, us-east-ny, us-mia), no multiplier.
- unverified: may bill active CPU/RAM rather than allocation. The pricing page lists per-vCPU-h / per-GB-h rates metered 'while a resource runs' without saying allocated vs used; the sandbox usage endpoint 'Returns measured runtime, provisioned vCPU time, active CPU, network traffic, estimated cost' (miosa.ai/docs/api-reference/sandboxes). Card keeps allocation (checked 2026-09-28).
- Idle auto-stop exists but is opt-in: idle_timeout_sec defaults to 0 (disabled); timeout_sec defaults to 3600 s (max 86400).
- Idle auto-stop exists but is off by default (must be configured), so the estimator does not apply idle suspension; set a pinned offer if you enable it.
- BYOC is fee-only + your cloud bill: "Your cloud provider bills you directly for the underlying compute, storage, and network — MIOSA only charges for orchestrating it, at 45% of the managed rate" (miosa.ai/pricing, 2026-09-28). Its row is the orchestration fee alone, flagged alt; add the AWS/GCP/Azure cost of the same VM.
- Disk is not sold separately for sandboxes: disk_size_mb "must match the selected canonical size" (miosa.ai/docs/api-reference/sandboxes, 2026-09-28); each size carries a fixed disk (sandbox medium 20 GB, Computer medium 50 GB) billed at 0.06147 credits/GB-h while running, folded into the size price. A bigger disk means a bigger size or a Computer, so disk_gib_month stays null.

## How this provider charges


MIOSA uses one credit rate card (1 credit = $0.01) for every product: sandboxes, desktop "Computers", deployments and
data. Plans never change the price. They set a **monthly minimum that is credited 1:1** and the caps and concurrency. The
regimes that change the bill are the plan minimum, running vs paused disk rates, the size preset (disk is billed while running), egress, and BYOC at 45%.

Rate card: vCPU **4.536 credits/h ($0.04536)**, RAM **1.458/GB-h ($0.01458)**, disk while running **0.06147/GB-h ($0.0006147)**,
cold storage while paused **0.00243/GB-h ($0.0000243, $0.01774/GB-month)**, egress **5/GB ($0.05)**, GPU **175/h ($1.75)**,
builds 2/min, LLM tokens at provider cost + 15%. The page rounds vCPU to $0.0454 and RAM to $0.0146.
Sandbox medium 4 vCPU / 8 GB / 20 GB = 18.144 + 11.664 + 1.2294 = **31.0374 credits = $0.310374/h** ($0.29808 without disk).

## Regime table

| Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|
| Developer | Default, no card | $0 fee. One-time credit. **Hard stop at $0**, no overage, no auto-recharge | $10 one-time (1,000 credits). 3 Computers, 10 Sandboxes, 2 concurrent | https://miosa.ai/pricing |
| Business | Production | $500/mo **minimum credited 1:1** (50,000 credits). Overage at the same rate with spend controls. Top-ups $1-$5,000 | 100 Computers, 250 Sandboxes, 50 concurrent. Cancel at end of period | https://miosa.ai/pricing |
| Enterprise | Request | $1,000/mo minimum credited 1:1 (100,000 credits). Same rates | 200 Computers, 500 Sandboxes, 200 concurrent. SSO/SCIM. Cancellation per agreement | https://miosa.ai/pricing |
| Per-computer caps | By plan | Size ceiling | Dev 2 vCPU/4 GB/10 GB, Business 8/16/40, Enterprise 16/32/80. **These were on the page earlier on 2026-09-28 and are gone on re-check** | raw/miosa_ai_pricing.txt vs raw/miosa_ai_pricing_recheck.txt |
| Sandbox sizes (running) | Sandbox API | Fixed presets, allocation, vCPU+RAM+disk per hour, metered continuously, ledger settles about every 5 min | xs 1/2/10 $0.0807, small 2/4/10 $0.1552, medium 4/8/20 $0.3104, large 8/16/40 $0.6207, xl 16/32/80 $1.2415 per h | https://miosa.ai/docs/api-reference/sandboxes |
| Computer sizes (running) | Desktop Computers (Xfce, Firefox) | Same rate, bigger disks | xs 10, small 20, medium 50, large 100, xl 200 GB disk. small = 16.13 credits/h (page calculator) | https://miosa.ai/docs/computers/overview |
| Paused / stopped | `pause`, timeout on a persistent sandbox, stopped Computer | No compute. Disk at the cold-storage rate | $0.0000243/GB-h = $0.01774/GB-month | https://miosa.ai/pricing |
| Timeout | `timeout_sec` 1-86,400 (default 3,600). `idle_timeout_sec` default 0 (off) | Persistent sandboxes **pause** at timeout. Non-persistent ones are destroyed | max 24 h per active session | https://miosa.ai/docs/api-reference/sandboxes |
| Egress | All outbound | Per GB | $0.05/GB, no free allowance published | https://miosa.ai/pricing |
| BYOC | Your AWS/GCP/Azure (gated private preview, needs /dev/kvm) | MIOSA orchestration fee = 45% of the managed rate. **Your cloud bill comes on top** | vCPU $0.0204, RAM $0.006561/GB-h, disk $0.00027662/GB-h, cold $0.00001094/GB-h, egress $0.0225/GB, GPU $0.7875/h | https://miosa.ai/pricing |
| GPU | Rate card only | $1.75/GPU-h (type unspecified). Managed GPU sandboxes are "planned", so today GPU only works via BYOC/OpenComputers | $1.75 | https://miosa.ai/pricing, /docs/known-limitations |
| Builds / tokens | Template builds, agent runs | Per build-minute. Tokens at cost + 15% | 2 credits/min | https://miosa.ai/pricing |
| Stale docs model | /docs/api-reference/credits | Old size-based credits (Small 4GB/1CPU 3/h, Medium 6, Large 12). Plans Free/Starter $29/Pro $79/Scale $199. Credits expire after 180 days | **Superseded by the pricing page, not used** | https://miosa.ai/docs/api-reference/credits |
| Regions | us-west-la, us-east-ny, us-mia | No multiplier | 1 | https://miosa.ai/docs/api-reference/regions |

## Gotchas

1. **The plan fee is a minimum, not an add-on.** Business at $500 costs max($500, usage). At low usage you pay $500 for less.
2. **Disk is billed while running** as part of the hourly rate. Computers carry 2-2.5× more disk than sandboxes of the same size, so a medium Computer costs $0.3288/h against $0.3104/h for a medium sandbox.
3. **Paused state is cheap but not free**: $0.01774/GB-month on the retained disk.
4. **Developer stops hard at $0** and allows only 2 concurrent workloads. The docs also cap tenants at 10 concurrent sandboxes by default ("contact support"), even though the Business page says 50.
5. **Per-computer size caps disappeared from the pricing page** during the day (2026-09-28). They may still be enforced.
6. **BYOC "45%" understates the total.** AWS/GCP/Azure bill you separately for the same VMs.
7. The **credits docs page is stale** (a different price model). Trust https://miosa.ai/pricing.
8. GB vs GiB is not specified. Billing granularity/minimum is unpublished ("metered continuously").

## Worked example

Workload: 4 vCPU / 8 GiB, 50 concurrent × 8 h/day × 22 days = **8,800 sandbox-hours**, 30% CPU, 50 GiB snapshots retained,
100 GiB egress. Medium sandboxes (4/8/20 GB) are paused between shifts.

- Compute + running disk: 8,800 × $0.310374 = **$2,731.29** (vCPU $1,596.67 + RAM $1,026.43 + disk $108.19). 30% CPU changes nothing (allocation).
- Paused state: 50 GiB × $0.01774 = **$0.89**.
- Egress: 100 × $0.05 = **$5.00**.
- Usage total: **$2,737.18**.

| Regime | Feasible? | Monthly total |
|---|---|---|
| Developer | No: 2 concurrent (and 2 vCPU cap per the earlier capture) | n/a |
| Business ($500 credited) | Yes, exactly 50 concurrent (the docs' default 10/tenant needs a support raise) | **$2,737.18** (fee absorbed) |
| Enterprise ($1,000 credited) | Yes | $2,737.18 |
| BYOC (orchestration only) | Private preview | 0.45 × 2,737.18 = **$1,231.73** + your own cloud bill |
| Anti-pattern: left running 24 h/day | Yes | 26,400 × 0.310374 + 5 = **$8,198.87** |