Why teams and developers run agents on Sandscale
Agents need a place to run code nobody has read
Running agent code on your own machines
Running it on Sandscale
See insideevery agent run
Claude Code runs under your policy through its hooks: each tool call is allowed, blocked or sent to a person. Codex and Gemini CLI report every step over OpenTelemetry. Pause or stop any session from the dashboard.
Policyand audit
Decide which templates, sizes, run times and agents each project may use, and which calls need a person first. Every sandbox, workspace, lease and change lands in a hash-chained audit log you can export.
Start free. Scale when your agents do.
(Every plan runs on the same isolated microVMs.)
Sandboxes, workspaces and build hosts all draw from your plan's monthly awake hours. A box counts only while it is awake: paused and idle time between tool calls is free. Pro gives one developer room to work; Team adds the controls a company needs; Enterprise adds your own cloud and your own keys.
| Free | Pro | Team | Enterprise | |
|---|---|---|---|---|
| Compute | ||||
| Members | 1 | 1 | 2 or more | Custom |
| Included awake vCPU-hours a month | 100 | 1,500 | 2,000 per member, pooled | Custom |
| Paused and idle time | Free | Free | Free | Free |
| Boxes running at once | 3 | 30 | 30, more on request | Custom |
| Largest box | Up to 2 vCPU | Up to 16 vCPU | Up to 16 vCPU, 32 on request | Up to 64 vCPU |
| Session length | Up to 4 h | Up to 24 h | Up to 24 h | Custom |
| Burst (flex compute) | No | Up to 4x the base size | Up to 8x the base size | Custom |
| Memory | 4 GiB per vCPU included | 4 GiB per vCPU included | 4 GiB per vCPU included | 4 GiB per vCPU included |
| Usage beyond your hours | No | $0.05 per awake vCPU-hour, on or off | $0.05 per awake vCPU-hour, on or off | Per agreement |
| Hour packs | No | $10 per 500 awake vCPU-hours | $10 per 500 awake vCPU-hours | Per agreement |
| Spend caps per task, agent, project and organization | No | ✓ | ✓ | ✓ |
| Dedicated cores | No | $0.04 per vCPU-hour | $0.04 per vCPU-hour | ✓ |
| Regions | US now, Europe coming soon | US now, Europe coming soon | US now, Europe coming soon | US now, Europe coming soon |
| Products | ||||
| Sandboxes | ✓ | ✓ | ✓ | ✓ |
| Pause and resume | ✓ | ✓ | ✓ | ✓ |
| Snapshots | ✓ | ✓ | ✓ | ✓ |
| Workspaces | One, sleeps after 30 min idle | ✓ | ✓ | ✓ |
| Build hosts, GitHub Actions runner | No | ✓ | ✓ | ✓ |
| Lab hosts | No | No | ✓ | ✓ |
| Add-ons (optional): always-on workspaces, build minutes | No | From $9 a month | From $9 a month | Per agreement |
| Agent Cloud | No | ✓ | ✓ | ✓ |
| Preview links | With your team | With your team or public | With your team or public | With your team or public |
| Run traces with OTLP export | ✓ | ✓ | ✓ | ✓ |
| Storage | ||||
| Included storage (disks, snapshots, templates, objects) | 5 GB | 50 GB | 100 GB per member, pooled | Custom |
| Snapshots and templates | First 10 GB free | First 10 GB free | First 10 GB free | Custom |
| Storage above your allowance | No | $0.15 per GB-month | $0.15 per GB-month | Per agreement |
| Network and security |
On paid plans you choose what happens when your included hours run out: turn usage beyond them on at $0.05 per awake vCPU-hour, add an hour pack of 500 awake vCPU-hours for $10, or keep it off and new boxes wait for next month. Running boxes are never stopped. Set spend caps per task, agent, project and organization.
Need bigger machines? Pro runs boxes of up to 16 vCPU, Team up to 32 on request, and Enterprise up to 64. Every box gets exactly the vCPUs and memory you picked, plus any burst you set, never more.
Everything a run needs
Start it, watch it, stop it. From your agent, your terminal or the dashboard.
Start, pause, snapshot and stop sandboxes from the API, the sandscale CLI or MCP. Every action lands in a tamper-evident audit log.
Sandboxes, workspaces, build hosts and lab hardware run on the same platform, managed from one dashboard and billed as one monthly plan.
Its own kernel per run, the network off until your allowlist opens it, secrets kept in Vault and never in prompts, and a kill switch for your organisation.
Connect over MCP from Claude Code, Codex or your own agent framework. Each agent gets its own box, and you see what each one is doing.