# Coasty

Official pricing: https://coasty.ai/pricing  
Category: agent-sandbox · Isolation: not published

## Pricing regimes (raw)

- **Linux running default** (sizes): Linux running default ? vCPU/? GiB $0.05/h
- **Windows running default** (sizes): Windows running default ? vCPU/? GiB $0.09/h

## Features

Yes: snapshots, pause/resume, persistent disk, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, browser, desktop GUI, computer-use API, browser automation API, desktop control API, browser + desktop control, residential IPs, SSH, public IPv4, open internet, BYOC, SOC 2, HIPAA, SSO, Windows, webhooks, scoped API keys, spend limits, MCP server, usage per sandbox (API), snapshots on demand, hosted agent API (their own agent)

No: memory snapshots, fork/clone, volumes, idle auto-stop, your own Docker/OCI image, nested virtualization, anti-bot stealth, CAPTCHA solving, preinstalled agents, HTTPS preview URLs, egress allowlist, static egress IP, self-hosting, open source, GPU, arm64, macOS, HTTP method/path egress rules, wake on request, live resize, live fork (no pause), memory fork, EU data residency, automatic snapshots, agent harness API, inbound access rules, secret proxy, secret proxy for any API, extra volumes, shared volumes

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

## Company

- Ownership: independent
- Stage: pre-seed [source](https://www.ycombinator.com/companies/coarena)
- Accelerators: Y Combinator S26 [source](https://www.ycombinator.com/companies/coarena)
- Total raised: unknown
- Last round: unknown (unknown)
- Investors: Y Combinator
- Revenue: unknown

coasty.ai (the sandbox/computer-use VM product this card covers) and YC S26 company 'CoArena' (coarena.ai, a crowdsourced computer-use benchmark) share the same two founders (Prateek Jannu, Nitish Kovuru) and the same corporate identity (LinkedIn 'coastyai', Crunchbase org 'coasty', contact founders@coasty.ai) per the YC company page. Founder bios state Coasty was built first ('Built Coasty to #1 on the OSWorld benchmark') and CoArena is a newer, separate product line from the same team, now the one formally in Y Combinator (Summer 2026 batch). No independent funding amount, round, or revenue figures for either product were found in the time available; Crunchbase and LinkedIn pages returned access-blocked (403/authwall) during research.

## Caveats

- Storage, snapshot GiB-month and public IPv4 prices are not published in the reviewed sources; null is not free.
- No guaranteed vCPU allocation published; browser-hour is not a verified 4-vCPU/8-GiB VM equivalent. The sizes card keeps unknown hardware null; do not coerce it to zero or auto-rank against resource-guaranteed VMs.
- Browser RAM is not published unless a specific mode overrides it.
- Public pricing headline says managed machines included, but detailed API docs and machine rate card explicitly charge runtime. Use detailed meter and flag conflict; do not assume step price includes rented VM hours.
- GET /v 1/machines/pricing requires authentication (401 observed); published-default card is available anonymously at /api/pricing. No actual account-specific tariff was verified.
- Stateful inference session is not a VM session and has no per-minute meter; managed machine and scheduled consumer execution are different objects.
- Consumer subscription credits are quota units with variable dollar values, never the API wallet $.01 conversion. Parked-machine hourly charge cannot be encoded as per-GiB snapshot rate.
- Marketing claims of benchmark superiority and comparison prices are not used as pricing evidence.
- Machine size re-checked 2026-09-28: docs expose cpu_cores 1-16, memory_gb 1-64, storage_gb 8-500 ("auto" default, capped by tier) but no default size; public price feed (coasty.ai/api/pricing) lists a flat 5 credits/h Linux, 9 credits/h Windows ($0.01/credit), 1 credit/h stopped. Whether the rate varies with size is only in the authenticated /v1/machines/pricing feed, so sizes stay unpublished (vcpu/ram null).
- Consumer subscriptions (separate consumer-credit balance, not the API wallet, and currently not offered for self-serve purchase): Starter $19/mo 200 credits, 1 machine; Plus $50/mo 600 credits, 2 machines; Unlimited $99/mo fair-use. Credit boosts (paid subscription required): 150/$19, 500/$49, 1,200/$99 (coasty.ai/api/pricing).
- Agent inference meters are billed on top of machine runtime: predict $0.05, session create $0.10, session predict $0.04, ground $0.03, managed run step $0.05 (v1 $0.08). Surcharges: trajectory screenshot +$0.02 each, high-res image +$0.01, v1 engine +$0.03/request, prompt >500 chars +$0.01 (coasty.ai/docs pricing).
- Machine backend varies (checked 2026-10-14, coasty.ai/docs/llms.txt "Choosing a backend"): auto sends desktop/Windows tasks to Daytona (AWS fallback) and headless tasks to Azure containers; provider can be pinned to aws, daytona or azure. Isolation is therefore left unknown, against the pricing page's "true VM-level isolation" claim.
- macOS is supported only as a caller-operated external machine (POST /v1/machines/external); managed machines are linux or windows.
- Positioning shift (checked 2026-10-14): the coasty.ai homepage now reads "Models, environments, data and evaluation for frontier AI labs" (title "Coasty — Advancing computer use") and links to CoArena; the YC listing is now CoArena (S26, ycombinator.com/companies/coarena). The "Coasty AI Employee" self-serve product survives only in coasty.ai/pricing structured data, while the machine API and rate card remain in the docs. The card stays agent-sandbox (a computer-use model API with rented machines); the self-serve offer may be winding down.

## How this provider charges


As of 2026-09-28. USD unless explicitly stated; public list prices unless a row is marked promotion, historical, or quote-only.

## Regime table

| Regime | When it applies | How billed | Numbers / terms | Source |
|---|---|---|---|---|
| Running Linux / Windows | Managed machine | Minute rounded down | $0.05/$0.09 per hour published defaults | https://coasty.ai/api/pricing |
| Stopped/suspended | Machine parked | Runtime keep-alive | $0.01/hour; creating/error/terminated free | https://coasty.ai/api/pricing |
| Snapshot | Managed VM | Per creation | $0.01, not $/GiB | https://coasty.ai/api/pricing |
| BYOK | Inference with own key | Platform inference exempt | Provider tokens + managed runtime still charged | https://coasty.ai/docs?section=pricing |
| Consumer subscriptions | Dashboard/schedules | Different credit balance | $19/200 cr; $50/600 cr; $99 Unlimited; boosts separate | https://coasty.ai/api/pricing |
| Synthetic test keys | Mock API/VMs | Not real compute | $0; maximum 5 mock machines | https://coasty.ai/docs/llms.txt |
| API pay as you go | Account plan | Monthly fee / commitment | $0; concurrency unpublished/custom; Prepaid USD wallet; $.20 provisioning balance gate, not fee. First $5 advertised. | https://coasty.ai/pricing |
| Developer API | Account plan | Monthly fee / commitment | $19; concurrency unpublished/custom; $25 monthly API credits; same $.05 managed step; consumer credits are separate. | https://coasty.ai/pricing |

## Gotchas

- Public pricing headline says managed machines included, but detailed API docs and machine rate card explicitly charge runtime. Use detailed meter and flag conflict; do not assume step price includes rented VM hours.

- GET /v 1/machines/pricing requires authentication (401 observed); published-default card is available anonymously at /api/pricing. No actual account-specific tariff was verified.

- Stateful inference session is not a VM session and has no per-minute meter; managed machine and scheduled consumer execution are different objects.

- Consumer subscription credits are quota units with variable dollar values, never the API wallet $.01 conversion. Parked-machine hourly charge cannot be encoded as per-GiB snapshot rate.

- Marketing claims of benchmark superiority and comparison prices are not used as pricing evidence.

- Storage, snapshot GiB-month and public IPv4 prices are not published in the reviewed sources; null is not free.

- No guaranteed vCPU allocation published; browser-hour is not a verified 4-vCPU/8-GiB VM equivalent. The sizes card keeps unknown hardware null; do not coerce it to zero or auto-rank against resource-guaranteed VMs.

- Browser RAM is not published unless a specific mode overrides it.

## Browser and desktop dimensions

```json
{
  "hour_usd": 0.05,
  "included_hours": null,
  "proxy_residential_gb_usd": null,
  "proxy_datacenter_gb_usd": null,
  "captcha_solve_usd": null,
  "stealth_surcharge_usd": null,
  "recording_storage_gib_month_usd": null,
  "recording_retention": null,
  "persistent_profiles": true,
  "resolution": "Inference HD surcharge >1280 px wide OR >720 px tall; VM resolution price tier not documented",
  "gpu_desktops": "General GPU-powered infrastructure claim; model, allocation and GPU rate not published",
  "os_options": [
    "linux",
    "windows"
  ],
  "browsers": [
    "Full desktop browser"
  ]
}
```

## Worked example

Requested: 4 vCPU / 8 GiB; 50 concurrent × 8 hours/day × 22 days = **8,800 runtime hours**, 1,100 eight-hour starts if recreated daily; 30% CPU; 50 GiB-month snapshots; 100 GiB egress.

Using published default Linux rates: **$440** for 8,800 running hours, or **$792** Windows. If all 50 are retained for a 730 h month, stopped hours=50×(730−176)=27,700, adding **$277**, so Linux **$717** / Windows **$1,069** before inference, snapshots and network. 1,100 created snapshots would add $11; 50 GiB storage does not imply 50 snapshots. Fifty live computers and 4/8 capacity need confirmation. Add $.05×managed_steps (or BYOK provider tokens), not 30% CPU.

## Sources

- https://coasty.ai/pricing

- https://coasty.ai/docs?section=pricing

- https://coasty.ai/docs/llms.txt

- https://coasty.ai/api/pricing

- https://coasty.ai/v1/machines/pricing