This repo contains a local launcher for a Lima VM that installs and runs the OpenClaw gateway inside Ubuntu 24.04, configured for DeepSeek and reachable from the macOS host through Lima port forwarding.

On a new Mac, install Lima first:

brew install lima

Then create the local config and start a VM:

scripts/lima-claw init
# edit lima-claw.env and set DEEPSEEK_API_KEY
scripts/lima-claw start openclaw
scripts/lima-claw chat openclaw "Say hello in one short sentence."

The default host URL is:

http://127.0.0.1:18789/

The gateway also enables the OpenAI-compatible HTTP chat endpoint:

curl -sS http://127.0.0.1:18789/v1/chat/completions \
  -H 'Content-Type: application/json' \
  --data '{"model":"openclaw","messages":[{"role":"user","content":"Say hello"}],"stream":false}'

scripts/lima-claw reads lima-claw.env by default. If only the old openclaw-lima.env exists, it will use that for compatibility. Config files are ignored by git because they contain the DeepSeek API key.

The default VM shape is intentionally modest for running several instances on a Mac:

INSTANCE_NAME=openclaw
CPUS=2
MEMORY=4GiB
DISK=30GiB
WORKSPACE_MOUNT=.
WORKSPACE_MOUNT_WRITABLE=false
GUEST_PORT=18789
HOST_PORT=18789
DEEPSEEK_MODEL=deepseek/deepseek-v4-flash
OPENCLAW_VERSION=latest
OPENCLAW_POST_INSTALL_COMMANDS=

By default the VM mounts only the repo directory, not your whole macOS home directory. Keep WORKSPACE_MOUNT_WRITABLE=false unless you explicitly want the VM to be able to edit the mounted host path.

Lima instance shape is fixed at creation time. If you change VM-level settings such as CPU, memory, disk, mount, or forwarded ports for an existing VM, run:

scripts/lima-claw destroy openclaw
scripts/lima-claw start openclaw

OPENCLAW_POST_INSTALL_COMMANDS runs inside the VM after Node/OpenClaw are installed and before the gateway starts. It is intended for non-interactive OpenClaw plugins or integration installers, for example:

OPENCLAW_POST_INSTALL_COMMANDS="openclaw plugins install '@example/openclaw-plugin'"

The script records a marker for the exact command string, so it runs again only when you change that config value.

For installers that print a QR code or ask for input, SSH into the VM instead:

scripts/lima-claw ssh openclaw
npx -y @tencent-weixin/openclaw-weixin-cli@latest install

The VM setup exposes OpenClaw's bundled Node toolchain in SSH shells, so node, npm, and npx are available after scripts/lima-claw start. It also installs the Ubuntu Go toolchain plus common build helpers, so go, make, and a C/C++ compiler are available in the VM. Go-installed binaries under $HOME/go/bin are added to the VM shell PATH.

For three concurrent VMs, create one config per instance and vary at least INSTANCE_NAME and HOST_PORT, for example 18789, 18790, and 18791.

scripts/lima-claw list                  # show Lima VMs
scripts/lima-claw start [name]          # create/start VM, configure OpenClaw, wait for /readyz
scripts/lima-claw resume [name]         # start an existing stopped VM and gateway
scripts/lima-claw restart [name]        # restart the OpenClaw gateway in a running VM
scripts/lima-claw health [name]         # check gateway /healthz and /readyz
scripts/lima-claw status [name]         # show Lima status plus /healthz and /readyz
scripts/lima-claw chat [name] "hello"   # send one OpenAI-compatible HTTP chat request
scripts/lima-claw ssh [name]            # SSH into the VM
scripts/lima-claw shell [name]          # open a limactl shell in the VM
scripts/lima-claw stop [name]           # stop the gateway process and Lima VM
scripts/lima-claw destroy [name]        # delete the Lima VM and generated state