Speedwave vs Claude Code
Claude Code is Anthropic’s coding agent. Speedwave runs it for you, inside a container and a VM, and adds the pieces a team needs to use it safely.
What plain Claude Code gives you
Section titled “What plain Claude Code gives you”Plain Claude Code is a terminal tool. You configure one model provider, run and secure any MCP server yourself, and approve each edit and command as Claude proposes it. It reads files anywhere your user account can read, not just the project you started it in. Packages it installs land on your machine and accumulate over time.
What Speedwave adds
Section titled “What Speedwave adds”| Capability | Plain Claude Code | With Speedwave |
|---|---|---|
| Interface | A terminal | Desktop chat app, plus the same CLI |
| Where Claude runs | On your machine | Hardened container, inside a VM (Lima on macOS, WSL2 on Windows) |
| Service tokens Speedwave can reach | Whatever is in its environment | None |
| Model provider | Configured by you | Anthropic, a local server, or OpenRouter, per project |
| MCP servers for integrations | You run and secure each one | Run for you, each in its own container |
| What Speedwave can read on disk | Anywhere your user can read | Only the active project and its container home |
| Approving each action | You approve every edit and command | Runs with permissions bypassed, no prompts |
| Installing packages for a task | Lands on your machine, accumulates | Inside the container, gone on rebuild |
| Data from integrations | Reaches the model as is | Sensitive values masked before the model sees them |
| Code review, including security | Ad hoc, on request | Bundled code-review skills, including a security checker |
| Claude Code version | Self-updates, whenever a new release ships | Pinned, updated with tested Speedwave releases |
Isolation is the base of the list: a container hardened to OWASP guidance, inside a VM as a further layer between Claude and your host. A SecurityCheck gate validates the setup before any container starts. Read the full model on the security pages.
Speedwave holds no tokens itself. Each integration runs in its own worker container with only its own credential, and every tool call routes through one hub that masks sensitive data before it reaches the model. Because every call passes through that single hub, its operations are auditable. Provider choice is per project; see LLM providers for how the switch works.
Plain Claude Code stops before every edit and command, waiting for you to approve it, on your own machine, with your full user account. Speedwave runs Claude with permission prompts bypassed, so it edits and runs commands without stopping to ask. That’s only safe because of the container around it: no capabilities beyond what the job needs, a read-only filesystem, no service tokens in reach, and network access scoped to the project. The container is what turns “skip all confirmations” from a risk into a shortcut.
Speedwave also pins the Claude Code version instead of letting it auto-update mid-session. The installer verifies the binary against a version-pinned manifest when it builds the container image, and each Speedwave release ships a Claude Code version that’s been tested against it. Standalone Claude Code updates itself whenever a new release is out; with Speedwave, updates arrive on your schedule, not Anthropic’s.
What Speedwave does not do
Section titled “What Speedwave does not do”Speedwave does not certify or guarantee GDPR, DORA, or other regulatory compliance. It has no built-in requirements-gathering or design feature. What to build and how to design it stays your responsibility, with Speedwave assisting rather than deciding.