# Sandscale > Isolated Firecracker microVMs for AI agents, with Vault, DLP, approvals and a tamper-evident audit log built in. Memory included. Sandscale runs the code AI agents write in sandboxes, workspaces, build hosts and lab hosts, each its own Firecracker microVM. Use it from the TypeScript and Python SDKs, the `sandscale` CLI, the REST API or the remote MCP server at mcp.sandscale.dev. Hosted in the US; Europe is coming soon. Sign up free at https://cloud.sandscale.dev/auth/signup. Every page below is also available as Markdown at the linked .md address, and pages are served as Markdown to requests with `Accept: text/markdown`. All docs in one file: https://sandscale.dev/llms-full.txt ## Docs - [Docs](https://sandscale.dev/docs.md): Sandscale developer docs. Start a sandbox for the code your agents write, give coding agents a persistent cloud computer, and keep every run governed and audited. - [Using the API](https://sandscale.dev/docs/api.md): How to call the Sandscale API, the hosts, request and response format, identifiers and where to find each endpoint. - [Authentication](https://sandscale.dev/docs/api/authentication.md): Authenticate to the Sandscale API with a personal API key or your own sign-in. Key scopes, expiry, rotation and revocation. - [Errors](https://sandscale.dev/docs/api/errors.md): How the Sandscale API reports a refused request, the error body, the status codes and the error codes you meet on many endpoints. - [Limits and retries](https://sandscale.dev/docs/api/limits.md): Rate limits, plan quotas and how to retry Sandscale API requests safely. - [Agent Cloud](https://sandscale.dev/docs/concepts/agent-cloud.md): Sandscale Agent Cloud gives coding tasks to agents running in your workspaces, lets you watch them live, and returns a branch and pull request. - [Approvals and policies](https://sandscale.dev/docs/concepts/approvals.md): Sandscale approval gates and policies on templates, sizes, run times and which agents may run. Approvals on every plan; policies on Team and Enterprise. - [Audit log](https://sandscale.dev/docs/concepts/audit.md): Sandscale keeps a hash-chained audit log of every action. Every plan can read it; Team and Enterprise can export it as evidence. - [Build hosts](https://sandscale.dev/docs/concepts/build-hosts.md): Sandscale build hosts are clean, leased build machines for agent tasks, released when the task ends. - [Lab hosts](https://sandscale.dev/docs/concepts/lab-hosts.md): Sandscale lab hosts connect your own Linux or macOS machines, with their hardware, so a run can claim them. Team and Enterprise. - [Object storage](https://sandscale.dev/docs/concepts/object-storage.md): Sandscale object storage is S3-compatible storage for files, datasets and build artifacts that outlive a run. United States now; Europe coming soon. - [Regions](https://sandscale.dev/docs/concepts/regions.md): Sandscale runs in the United States now; Europe is coming soon. One account, with your workloads and their data kept in the region they run in. API hosts api.sandscale.dev and api.us.sandscale.dev. - [Sandboxes](https://sandscale.dev/docs/concepts/sandboxes.md): A Sandscale sandbox is a Firecracker microVM for the code your agents write, with its own kernel and the network off by default. - [Teams and SSO](https://sandscale.dev/docs/concepts/teams-and-sso.md): Sandscale organizations, invitations, roles, groups and SSO on Team; SCIM, verified domains and custom roles on Enterprise; passkeys and multi-factor sign-in on every plan. - [Vault](https://sandscale.dev/docs/concepts/vault.md): Sandscale Vault keeps secrets out of prompts and logs. Secrets are referenced, injected at run time and never placed in prompts or logs. - [Workspaces](https://sandscale.dev/docs/concepts/workspaces.md): A Sandscale workspace is a persistent cloud computer for coding agents and people. Files survive, you open it in the browser or over SSH, and it stops when idle. - [Lease a build host with the API](https://sandscale.dev/docs/guides/build-hosts-api.md): Lease a build host from a pool, connect over SSH with a short-lived certificate, run and record commands, renew and release the lease. - [Connect Claude or ChatGPT](https://sandscale.dev/docs/guides/connect-claude-chatgpt.md): Add mcp.sandscale.dev as a remote MCP connector in Claude, ChatGPT or any remote MCP client and run sandboxes from chat, with your own sign-in. - [Store files in object storage](https://sandscale.dev/docs/guides/object-storage.md): Create a Sandscale object storage bucket and an S3 access key, find your region's S3 endpoint, and use standard S3 tools and SDKs. - [Run code in a sandbox with the API](https://sandscale.dev/docs/guides/sandboxes-api.md): Start a sandbox, run commands, move files in and out and stop it, with the Sandscale HTTP API. - [Pause, resume and snapshot a sandbox](https://sandscale.dev/docs/guides/snapshots.md): Pause a Sandscale sandbox and resume it later, save its /workspace as a snapshot, and start new sandboxes from it. - [Give a run a secret](https://sandscale.dev/docs/guides/vault-secrets.md): Store a secret in Sandscale Vault once and give it to a sandbox as an environment variable, without the value passing through your code, the API answer or a log. - [Work in a cloud workspace](https://sandscale.dev/docs/guides/workspaces.md): Create a Sandscale workspace from a template, open it over SSH or in the browser, forward ports, set a schedule and stop it, with the CLI or the SDKs. - [Quickstart](https://sandscale.dev/docs/quickstart.md): Sign up for Sandscale, create an API key, run a command in a sandbox with the TypeScript SDK, connect Claude Code, Codex or Cursor over MCP, and pick a region. - [SDKs](https://sandscale.dev/docs/sdks.md): The Sandscale SDKs for TypeScript and Python: install, authenticate, pick a region, start a sandbox and run a command, handle errors and retries. - [Python SDK](https://sandscale.dev/docs/sdks/python.md): Every method and option of the Sandscale Python SDK (sandscale 0.2.0), generated from its source. - [TypeScript SDK](https://sandscale.dev/docs/sdks/typescript.md): Every method and option of the Sandscale TypeScript SDK (@sandscale/sdk 0.2.0), generated from its source. ## Reference - [API reference](https://sandscale.dev/docs/reference/api.md): Every Sandscale API endpoint, by area, with parameters, request and response fields and error codes, generated from the server's code. - [Roles and permissions](https://sandscale.dev/docs/reference/api/access.md): The Roles and permissions endpoints of the Sandscale API, generated from the server's code. - [Your account and API keys](https://sandscale.dev/docs/reference/api/account.md): The Your account and API keys endpoints of the Sandscale API, generated from the server's code. - [Agent Cloud and agent tasks](https://sandscale.dev/docs/reference/api/agent-cloud.md): The Agent Cloud and agent tasks endpoints of the Sandscale API, generated from the server's code. - [Approvals, policies and the kill switch](https://sandscale.dev/docs/reference/api/approvals-policies.md): The Approvals, policies and the kill switch endpoints of the Sandscale API, generated from the server's code. - [Audit log](https://sandscale.dev/docs/reference/api/audit.md): The Audit log endpoints of the Sandscale API, generated from the server's code. - [Build hosts and lab hosts](https://sandscale.dev/docs/reference/api/build-hosts.md): The Build hosts and lab hosts endpoints of the Sandscale API, generated from the server's code. - [Object storage](https://sandscale.dev/docs/reference/api/object-storage.md): The Object storage endpoints of the Sandscale API, generated from the server's code. - [Organization, members and teams](https://sandscale.dev/docs/reference/api/organization.md): The Organization, members and teams endpoints of the Sandscale API, generated from the server's code. - [Usage, projects, quotas and regions](https://sandscale.dev/docs/reference/api/projects-quotas.md): The Usage, projects, quotas and regions endpoints of the Sandscale API, generated from the server's code. - [Sandboxes](https://sandscale.dev/docs/reference/api/sandboxes.md): The Sandboxes endpoints of the Sandscale API, generated from the server's code. - [Sign-in and OAuth](https://sandscale.dev/docs/reference/api/sign-in.md): The Sign-in and OAuth endpoints of the Sandscale API, generated from the server's code. - [SSO and SCIM](https://sandscale.dev/docs/reference/api/sso.md): The SSO and SCIM endpoints of the Sandscale API, generated from the server's code. - [Vault and secrets](https://sandscale.dev/docs/reference/api/vault.md): The Vault and secrets endpoints of the Sandscale API, generated from the server's code. - [Workspaces](https://sandscale.dev/docs/reference/api/workspaces.md): The Workspaces endpoints of the Sandscale API, generated from the server's code. - [CLI reference](https://sandscale.dev/docs/reference/cli.md): Every sandscale CLI command, with its usage, options and examples, generated from the CLI itself. - [sandscale build](https://sandscale.dev/docs/reference/cli/build.md): sandscale build: Build hosts and lab hosts: pools, your leases, runs and their logs. Usage, options and examples, generated from the CLI. - [sandscale help](https://sandscale.dev/docs/reference/cli/help.md): sandscale help: Help for a command. Usage, options and examples, generated from the CLI. - [sandscale login](https://sandscale.dev/docs/reference/cli/login.md): sandscale login: Sign in to Sandscale (browser with PKCE; --device for a code on another device). Usage, options and examples, generated from the CLI. - [sandscale logout](https://sandscale.dev/docs/reference/cli/logout.md): sandscale logout: Sign out (revokes the session and deletes the local tokens). Usage, options and examples, generated from the CLI. - [sandscale logs](https://sandscale.dev/docs/reference/cli/logs.md): sandscale logs: Logs of a workspace, a sandbox or a build host run. Usage, options and examples, generated from the CLI. - [sandscale mcp](https://sandscale.dev/docs/reference/cli/mcp.md): sandscale mcp: An MCP server (stdio) with your workspaces, sandboxes and build hosts as tools. Usage, options and examples, generated from the CLI. - [sandscale sandbox](https://sandscale.dev/docs/reference/cli/sandbox.md): sandscale sandbox: Sandboxes. Usage, options and examples, generated from the CLI. - [sandscale vault](https://sandscale.dev/docs/reference/cli/vault.md): sandscale vault: Sandscale Vault: KV secrets, policies, leases, transit, mounts and audit devices. Usage, options and examples, generated from the CLI. - [sandscale version](https://sandscale.dev/docs/reference/cli/version.md): sandscale version: Print the version. Usage, options and examples, generated from the CLI. - [sandscale whoami](https://sandscale.dev/docs/reference/cli/whoami.md): sandscale whoami: Who is signed in. Usage, options and examples, generated from the CLI. - [sandscale workspace](https://sandscale.dev/docs/reference/cli/workspace.md): sandscale workspace: Cloud workspaces. Usage, options and examples, generated from the CLI. - [MCP tools](https://sandscale.dev/docs/reference/mcp.md): The tools sandscale mcp gives your coding agent: workspaces, sandboxes and build hosts, generated from the MCP server's tool list. - [Plans and limits](https://sandscale.dev/docs/reference/plans.md): Sandscale plans, prices, included awake vCPU-hours, limits and features. ## Product - [Agent Cloud](https://sandscale.dev/agent-cloud.md): Give coding tasks to agents running in your workspaces, watch them live, and get a branch and pull request back. A lead agent can run the team through one MCP address. - [Build hosts](https://sandscale.dev/build-hosts.md): Lease a large, clean Linux machine for the heavy part of an agent’s work. Access and secrets end with the lease and the host is wiped. - [Leases](https://sandscale.dev/build-hosts/leases.md): Your agent leases a large, clean host from a pool for the heavy part of its work. Access and secrets end with the lease, and the host is wiped. - [CLI](https://sandscale.dev/cli.md): The sandscale command line for macOS, Linux and Windows, coming soon: sandboxes, workspaces, files, snapshots, SSH, tunnels and MCP. - [Integrations](https://sandscale.dev/integrations.md): Connect coding agents to Sandscale over MCP, script it with the CLI, or call the API. SSO with SAML or OpenID Connect, SCIM on Enterprise, and workers in your own AWS account on Enterprise. - [Lab hosts](https://sandscale.dev/lab-hosts.md): Connect your own Linux and macOS machines with security keys, TPMs, debug probes and boards. Runs claim a device, a person approves when policy says so, and it is released when the run ends. - [Device policy](https://sandscale.dev/lab-hosts/device-policy.md): Write down which devices each project may claim, which claims need a person to sign off, and who that person is. Anything you didn’t allow is refused. - [Object storage](https://sandscale.dev/object-storage.md): S3-compatible object storage for files, datasets and build artifacts that outlive a run. Works with standard S3 tools and SDKs. US now, Europe coming soon. - [Regions](https://sandscale.dev/regions.md): Where Sandscale runs: US East (N. Virginia) on AWS now, Europe (Frankfurt) coming soon, or your own AWS account on Enterprise. What stays in your region and what is held in the US. - [Europe region](https://sandscale.dev/regions/eu.md): Sandscale in Europe (Frankfurt, Germany) on AWS eu-central-1: what runs there, what stays there, and the rules organisations there work under. - [United States region](https://sandscale.dev/regions/us-east.md): Sandscale in the United States (N. Virginia, United States) on AWS us-east-1: what runs there, what stays there, and the rules organisations there work under. - [Sandboxes](https://sandscale.dev/sandboxes.md): A Firecracker microVM for every agent run, with no network by default. Every prompt and tool call recorded and checked against your policy. - [Files in and out](https://sandscale.dev/sandboxes/files.md): Move a source tree into a sandbox and pull reports, builds and logs back out, up to 1 GiB per file. - [Live output](https://sandscale.dev/sandboxes/live-output.md): Run a command in a sandbox and read stdout and stderr as they happen, then the exit code, from the CLI, the API or your agent. - [MCP server](https://sandscale.dev/sandboxes/mcp.md): Run sandscale mcp and Claude Code, Codex, Cursor or any MCP tool can create a sandbox, run commands in it, move files and throw it away, every call on the record. - [Pause and resume](https://sandscale.dev/sandboxes/pause-and-resume.md): Pausing keeps a sandbox’s memory and files. Resuming brings it back as it was, so an agent can stop for the night and carry on in the morning. - [Snapshots](https://sandscale.dev/sandboxes/snapshots.md): Save a sandbox after the slow part, like installing dependencies, and start new sandboxes from that point instead of from scratch. - [SSH certificates](https://sandscale.dev/sandboxes/ssh.md): Open a shell in a sandbox or workspace with a certificate that expires in minutes, issued to you and recorded in the audit log. - [Authenticated tunnels](https://sandscale.dev/sandboxes/tunnels.md): See the app or API your agent started inside a sandbox, from your own machine, through a tunnel that only you can open. - [Warm pools](https://sandscale.dev/sandboxes/warm-pools.md): A warm pool keeps sandboxes from a template already booted, so when your agent asks for one it gets a running microVM straight away. - [Templates](https://sandscale.dev/templates.md): Sandbox templates pinned by image digest: every sandbox started from a template gets exactly the same bytes. Start from Python, Node.js, Go or Rust, or your own image. - [Workspaces](https://sandscale.dev/workspaces.md): A persistent cloud machine for Claude Code, Codex or Gemini CLI. Close your laptop and the agent keeps working; open the same session from any device. - [Editors and devices](https://sandscale.dev/workspaces/editors.md): Open your workspace in the browser, or point VS Code, Cursor or a terminal at it. The agent’s session lives there, so you can close the laptop and pick it up anywhere. - [Schedules and idle sleep](https://sandscale.dev/workspaces/schedules.md): Start and stop workspaces on a schedule, and let them sleep when nobody uses them. The disk stays; hours count only while they run. ## Pricing and programs - [Open source](https://sandscale.dev/open-source.md): Sandscale Pro free for 6 months for maintainers of active open-source projects: sandboxes, workspaces and build hosts for your CI and agents. - [Pricing](https://sandscale.dev/pricing.md): Sandscale plans: Free, Pro $20 a month, Team $40 per member a month, and Enterprise. Included awake vCPU-hours every month; paused and idle time is free. - [Startups](https://sandscale.dev/startups.md): Sandscale Team free for 6 months for startups: monthly credits cover your Team subscription. ## Security - [Security](https://sandscale.dev/security.md): How Sandscale contains agent code and proves it: a Firecracker microVM per run, no network by default, Vault, DLP on prompts and tool calls, human approvals, SSO and SCIM, a hash-chained audit log and compliance evidence. - [Agent policy](https://sandscale.dev/security/agent-policy.md): Claude Code runs under your policy through its hooks: each prompt is checked, and each tool call is allowed, blocked or sent to a person. Codex and Gemini CLI report every step. - [Approvals](https://sandscale.dev/security/approvals.md): When policy says a call needs sign-off, the agent waits while a person approves or denies it. An approved call runs once. - [Audit log](https://sandscale.dev/security/audit-log.md): Every sandbox, command, approval, secret and change is written to an audit log where each entry is chained to the one before it, so an edit shows. - [Vulnerability disclosure](https://sandscale.dev/security/disclosure.md): How to report a security vulnerability in Sandscale, what to include, and what happens after you report it. - [DLP on prompts and tool calls](https://sandscale.dev/security/dlp.md): On every plan, prompts and MCP tool calls are scanned before they go anywhere, so a credential pasted into a prompt never reaches the model. - [Evidence export](https://sandscale.dev/security/evidence-export.md): Every audit entry for a period, its hash chain checked, packed into one file you hand to your assessor. - [Roles](https://sandscale.dev/security/roles.md): Give people Viewer, User, Editor or Manager on each sandbox, template and project. Whoever starts a sandbox owns it. - [SSO and SCIM](https://sandscale.dev/security/sso-and-scim.md): Sign in with SAML or OpenID Connect through Microsoft Entra ID, Okta, Google Workspace or any standard identity provider, and keep members in step with your directory over SCIM. - [Vault](https://sandscale.dev/security/vault.md): Keys and tokens live in Vault and reach a sandbox, workspace or build host in memory, for as long as the run lasts. - [Signed webhooks](https://sandscale.dev/security/webhooks.md): Send sandbox events to your own endpoints, signed so you can check they came from Sandscale, and retried if your endpoint is down. - [Your own cloud](https://sandscale.dev/security/your-own-cloud.md): Sandscale workers can run in your own AWS account, through a role you create with an external ID. No long-lived keys, and your data stays in your account. ## Blog - [Blog](https://sandscale.dev/blog.md): Sandscale news, releases and how-tos, written plainly. - [Awake-time billing for agent sandboxes: what you pay for and what you don't](https://sandscale.dev/blog/awake-time-billing-agent-sandboxes.md): How Sandscale bills agent sandboxes by awake vCPU-hours, with idle time free. Includes a worked example with the arithmetic shown. - [Running Claude Code, Codex and Gemini CLI under policy on Sandscale](https://sandscale.dev/blog/coding-agents-under-policy-sandscale.md): Two ways to run coding agents on Sandscale, what policy enforces for Claude Code versus what it only records for Codex and Gemini CLI, and the gaps. - [Containers, gVisor or Firecracker: how to isolate code your AI agent wrote](https://sandscale.dev/blog/isolating-ai-agent-code-containers-gvisor-firecracker.md): A plain comparison of containers, gVisor and Firecracker microVMs for running untrusted agent code: threat model, how each works, and what each costs. - [Keeping secrets out of agent prompts: how Vault and DLP fit together](https://sandscale.dev/blog/secrets-out-of-agent-prompts-vault-dlp.md): How a secret leaks from an AI agent run, what Sandscale Vault and DLP each cover, and the paths neither closes. Documented behavior only. ## Company - [Brand](https://sandscale.dev/brand.md): The Sandscale mark, colours, type and how to use them. - [Changelog](https://sandscale.dev/changelog.md): Sandscale releases and changes, written plainly. - [About](https://sandscale.dev/company.md): Sandscale builds isolated compute for AI agents: sandboxes, workspaces, build hosts and lab hosts. Operated by Sandscale, Inc., a XXX company. - [Contact](https://sandscale.dev/contact.md): Every Sandscale inbox, read by the team: sales, Enterprise, billing, support, developers, security, abuse, privacy and legal. - [FAQ](https://sandscale.dev/faq.md): Straight answers about Sandscale: sandboxes, workspaces, build and lab hosts, isolation, speed, plans and regions. - [Status](https://sandscale.dev/status.md): The Sandscale status page goes live at launch. Hosted compute runs in US East. Europe is coming soon. - [Support](https://sandscale.dev/support.md): How to get help with Sandscale: what to send, and how support works on each plan. ## Optional - [Sandscale: sandboxes for AI agents](https://sandscale.dev/index.md): Isolated compute for AI agents. Firecracker microVM sandboxes, workspaces and build machines, started in one call. - [Acceptable use](https://sandscale.dev/acceptable-use.md): What you may and may not run on Sandscale sandboxes, workspaces, build hosts and lab hosts. - [Legal](https://sandscale.dev/legal.md): Sandscale policies: terms of service, privacy, cookies, data processing, subprocessors, service levels, acceptable use, export control and accessibility. - [Accessibility](https://sandscale.dev/legal/accessibility.md): The accessibility statement for the Sandscale website. - [Company information](https://sandscale.dev/legal/company-information.md): Registered details of Sandscale, Inc., a XXX company, and its parent. - [Cookie policy](https://sandscale.dev/legal/cookies.md): The cookies and browser storage the Sandscale website uses: none, apart from a remembered appearance setting. - [Data processing](https://sandscale.dev/legal/data-processing.md): A summary of the Sandscale data processing addendum: roles, commitments, where data is processed, subprocessors and transfers. - [Export control](https://sandscale.dev/legal/export-control.md): Export control and sanctions compliance for Sandscale. - [Privacy notice](https://sandscale.dev/legal/privacy.md): How Sandscale collects, uses and protects personal information, who we share it with, how long we keep it, and your rights. - [Service levels](https://sandscale.dev/legal/service-levels.md): Sandscale availability and support commitments, and how they apply during early access. - [Subprocessors](https://sandscale.dev/legal/subprocessors.md): The subprocessors Sandscale uses to provide the service: what each does and where. - [Terms of service](https://sandscale.dev/legal/terms.md): The terms of service for Sandscale: accounts, plans and billing, your content, suspension, liability and governing law.