# Droneship

> Cloud sandboxes for coding agents. Launch Claude Code or Codex on an
> isolated cloud machine with one command, and give it real work
> without giving it your credentials.

Droneship is built by Highflame (https://highflame.com). It runs on macOS and
Linux, against git repos whose origin is on GitHub.

## Get started

    curl -fsSL https://droneship.run/install | sh
    droneship run claude

- [Quickstart](https://droneship.run/docs/start/quickstart/): install the CLI, sign in, and run Claude Code on your repo in a cloud sandbox
- [Concepts](https://droneship.run/docs/start/concepts/): sandboxes, harnesses, workspaces, prompt runs and projects
- [Docs home](https://droneship.run/docs/)

## What it is

Each run gets its own sandbox with its own kernel boundary (gVisor), its own
clone of your repo and its own short-lived identity. The image already carries
the common toolchain (Python, Node and Go; git, gh, npm, pnpm, yarn, pip, uv,
make, jq, ripgrep and curl). Each repo has a saved workspace, so when a
sandbox stops, the next run picks up its files and the agent's past
conversations. Sandboxes stop after 30 minutes with nobody connected and at an
8 hour cap.

Agents: Claude Code and OpenAI Codex CLI. Model calls bill to your own Claude
or ChatGPT plan, or to your own API key (--byok, or a key saved on the
console's Settings page).

## Security model

- **Egress**: an allowlist decides which hosts a sandbox can reach.
- **Agent gateway**: MCP and model calls leave through a gateway that refuses
  any MCP tool you have not turned on and checks model calls with Highflame's
  guardrails.
- **Credentials**: keys for connected services stay outside the sandbox and
  are added to allowed requests on the way out. Claude Code's model
  credential is held by a sidecar outside the sandbox.
- **Activity**: what the agent did in each task is recorded, along with what
  each sandbox spent on model calls.

- [Security overview](https://droneship.run/docs/security/)
- [Isolation and identity](https://droneship.run/docs/security/isolation-and-identity/)
- [Egress](https://droneship.run/docs/security/egress/)
- [Tools and MCP](https://droneship.run/docs/security/tools-and-mcp/)

## More

- [Harnesses and models](https://droneship.run/docs/guides/harnesses-and-models/)
- [Workspaces](https://droneship.run/docs/guides/workspaces/)
- [Run agents in parallel](https://droneship.run/docs/workflows/parallel/)
- [FAQ](https://droneship.run/docs/reference/more/faq/)
- [CLI reference](https://droneship.run/docs/reference/)

## Legal

- [Privacy Policy](https://www.highflame.com/privacy-policy)
- [Terms of Use](https://www.highflame.com/terms-of-use)