# PandaStack

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

## Pricing regimes (raw)

- **Hosted8-vCPU burstable template; active CPU and resident RAM** (resource): $0.054/vCPU-h, $0.0162/GiB-h

## 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, root, browser, code interpreter, preinstalled agents, HTTPS preview URLs, open internet, self-hosting, BYOC, open source, SSO, Python, Node.js, Go, wake on request, live fork (no pause), memory fork, webhooks, 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, SSH, public IPv4, egress allowlist, static egress IP, SOC 2, HIPAA, 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/pandastack.json → feature_evidence

## Caveats

- Every first-party template is pinned to 8 vCPUs, not selectable 4; RAM is baked per template. Live API lists Team CPU quota 16 and Enterprise 64, but quota does not prove launchable >8 template cores.
- Engine bumps 4-vCPU request to mandatory 8 and keeps CPU utilization fraction. Equal absolute CPU work requires adjusting utilization from 30% of 4 to 15% of 8.
- Live API Team SAML differs from pricing page Enterprise-only SSO presentation; confirm plan entitlement.
- Pricing FAQ says hibernated state storage currently included/free; lifecycle page says hibernated storage costs. Card follows explicit pricing FAQ but treats this as a documented conflict.
- Plain pause freezes CPU but retains RAM on the host; RAM billing while merely paused is not clearly specified. Do not equate pause and full hibernation.
- Volume pricing says overage beyond quota, but quota docs also describe hard HTTP 429 caps. Confirm how to buy over-quota storage.
- Persistent volumes are host-local; readonly multi-attach supported, writable attach exclusive. Cross-host staging is not yet shipped; not an S3-backed globally durable volume.
- Annual 17% discount advertised but credit treatment not verified; only monthly plans auto-ranked.
- Performance claims are vendor claims, not independent measurements.
- Disk (verified 2026-09-28): each template ships a fixed rootfs (base 2 GB, agent 6 GB, claude-agent 8 GB, browser 4 GB; docs/templates/overview) with no separate rootfs charge; included_disk_gib = 2 (base). Extra space comes from persistent volumes: plan quota included (Free 1 GiB, Pro 10 x 10 GiB = 100 GiB, Team 50 x 50 GiB = 2.5 TiB), then $0.15/GiB-month (pricing page, api.pandastack.ai/v1/pricing volume_usd_per_gib_month). The engine charges $0.15 on all disk beyond 2 GiB, so it overstates cost for disk that fits inside the plan's volume quota.
- SSH (verified 2026-09-28): 'pandastack sandbox ssh <id>' opens an interactive PTY over WebSocket (docs/sandboxes/terminal); no SSH protocol or SSH keys are documented, so standard ssh/scp/VS Code Remote-SSH clients don't work. ssh = false.
- Desktop (verified 2026-09-28): the 8 official templates (base, code-interpreter, agent, claude-agent, browser, 3 postgres) include no desktop/VNC/GUI template; the browser template is headless Chromium. desktop = false (a custom image could add one; unverified).
- Hourly create caps per workspace: Free 60, Pro 600, Team 6,000 (Enterprise custom). Each returns HTTP 429 'hourly create limit exceeded'. Forks, restored snapshots and template launches each count as a sandbox. Bursty fan-out beyond 600 creates/h forces Team.
- Non-persistent sandboxes are killed at TTL (default 1 h, refreshed by API activity; paid TTL cap Pro 7 d / Team 30 d idle timeout). Only persistent:true sandboxes are hibernated by the 5-min idle sweeper.

## How this provider charges


As of 2026-09-28. USD public list rates, independently matched between the [pricing page](https://www.pandastack.ai/pricing/), [billing docs](https://docs.pandastack.ai/docs/getting-started/billing/) and [live rate-card API](https://api.pandastack.ai/v1/pricing). Discovery was a [July 8 2026 comparison article](https://www.pandastack.ai/blog/best-e2b-alternatives-2026/), not proof of the exact original launch date.

| Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|
| Running | Hosted sandbox, app or database | Active CPU-seconds plus resident working-set GiB-seconds | $0.054/active-vCPU-hour; $0.0162/working-set-GiB-hour; per-second | [Live API](https://api.pandastack.ai/v1/pricing) |
| Burstable | All first-party templates | Same usage rate, no separately priced burst credits found | 8 exposed vCPU; Free up to2 sustained cores, Pro/Team up to8 with fair-share contention | [Templates](https://docs.pandastack.ai/docs/templates/overview/), [pricing](https://www.pandastack.ai/pricing/) |
| Paused | Guest CPU frozen on host | CPU inactive; RAM billing unclear | Do not apply zero-cost full-hibernation assumption | [Lifecycle](https://docs.pandastack.ai/docs/sandboxes/lifecycle/) |
| Hibernated | Memory+disk saved and VM released | No CPU/RAM billing; state currently included per pricing FAQ | $0 compute; $0 retained hibernation state on current pricing page | [Pricing FAQ](https://www.pandastack.ai/pricing/), [lifecycle](https://docs.pandastack.ai/docs/sandboxes/lifecycle/) |
| Free plan | 5 concurrent, max4GiB RAM,1-hour lifetime | Monthly grant, then pause; no surprise overage | $0 base + $5.40 credit/month;60 creates/hour | [API](https://api.pandastack.ai/v1/pricing) |
| Pro plan | Up to50 concurrent,16GiB RAM,600 creates/hour | Fee converts to usage credit | $20/month minimum spend; not $20 added to full compute bill | [API](https://api.pandastack.ai/v1/pricing) |
| Team plan | Up to500 concurrent,64GiB RAM,6,000 creates/hour | Fee converts to usage credit | $99/month minimum spend | [API](https://api.pandastack.ai/v1/pricing) |
| Volume storage | Provisioned block volumes above account plan quota | GiB-month | $0.15; included Free1GiB, Pro100GiB, Team2,500GiB | [Volume docs](https://docs.pandastack.ai/docs/volumes/overview/) |
| Network | Ingress and egress | Not billed subject to fair use | $0/GB | [API](https://api.pandastack.ai/v1/pricing), [pricing](https://www.pandastack.ai/pricing/) |
| Enterprise / self-host | Dedicated/BYOC contract or Apache2 software on own machines | Negotiated / external infrastructure | Unknown, not free hosted compute | [Pricing](https://www.pandastack.ai/pricing/), [self-host](https://docs.pandastack.ai/docs/self-host/overview/) |

## Gotchas

- RAM is fixed in template; **all first-party templates expose8vCPU**. Team's API quota of16vCPU and Enterprise's64 do not prove a supported larger create size. Use conservative actual template limits. [Templates](https://docs.pandastack.ai/docs/templates/overview/).
- CPU and RAM both track use, not just CPU. Need resident GiB over time; 30% CPU gives no memory-utilization assumption. [Billing](https://docs.pandastack.ai/docs/getting-started/billing/).
- Hibernation pricing conflicts with lifecycle prose saying retained state costs storage. The explicit pricing FAQ currently says included. Recheck before relying on free retained state. [Pricing](https://www.pandastack.ai/pricing/), [lifecycle](https://docs.pandastack.ai/docs/sandboxes/lifecycle/).
- Plain fork is filesystem-only; fork-tree preserves RAM+disk. Warm single-child fork brings the parent down. [Fork docs](https://docs.pandastack.ai/docs/sandboxes/snapshots-and-forks/).
- Volumes are **host-local**, readonly multi-attach or exclusive writer. Quota docs describe hard limits despite price-page overage language. Cross-host staging not shipped. [Volumes](https://docs.pandastack.ai/docs/volumes/overview/).
- Current official compliance docs expressly disclaim SOC2, ISO27001 and HIPAA attestations. SAML plan presentation also differs between live API and pricing page. [Compliance](https://docs.pandastack.ai/docs/security/compliance/).

## Required worked example

4 vCPU /8GiB requested;50 concurrent ×8h/day ×22days = **8,800 sandbox-hours**;30%CPU;50GiB-month hibernation state;100GiB egress. Pro permits50 and has800GiB account memory plus400vCPU capacity. Build an8GiB template. Mandatory8vCPU affects utilization interpretation:

1. **Same absolute work as 4vCPU at30%** =1.2 active cores, or15% of the8-core template. Assume all8GiB remain resident (RAM utilization not supplied):
   - CPU `8,800 × 4 × .30 × .054 = $570.24`.
   - RAM `8,800 × 8 × .0162 = $1,140.48`.
   - Total compute **$1,710.72**. Pro credit-converting minimum: `max($20,$1,710.72) = $1,710.72`.
2. **Engine sizing convention** bumps the allocation to8 cores and retains30%utilization: CPU **$1,140.48**, RAM **$1,140.48**, subtotal **$2,280.96**. This represents twice the absolute CPU work of case1.

Current pricing FAQ gives50GiB hibernation state $0 and100GiB egress $0, subject to stated conflict/fair-use caveats. No one-time/Free-plan grant applied to Pro. Unknown startup/minimum charges and capacity availability can still affect a real invoice. Optional persistent-volume charges are separate; do not turn all retained VM state into provisioned volume storage.