ajeetraina777/sbx-grok-kits repository overview

⁠Grok CLI kit for Docker Sandboxes

A standalone Docker Sandboxes⁠ kit (kind: mixin) that adds the Grok CLI⁠ (@vibe-kit/grok-cli) to any sandbox agent. Grok CLI is an OpenAI-compatible client, so this image ships in three backend flavors, one per tag.

Source and full docs: https://github.com/ajeetraina/sbx-kits-grok⁠

⁠Image tags

xAI is the default because Grok CLI is, first and foremost, a Grok client. The DMR tag is the zero-key path (it runs an open-weights model locally through the same CLI); the OpenAI tag reuses the CLI's OpenAI compatibility.

⁠Quick start

xAI (default). Store your key once with sbx, then run:

sbx secret set-custom -g --host api.x.ai --env GROK_API_KEY \
  --placeholder "xai-{rand}" --value "$XAI_API_KEY"
sbx run --kit docker.io/ajeetraina777/sbx-grok-kits:latest claude

Local (DMR). Enable Docker Model Runner and pull the model on the host - no key:

docker model pull ai/gemma3
sbx run --kit docker.io/ajeetraina777/sbx-grok-kits:dmr claude

OpenAI:

sbx secret set-custom -g --host api.openai.com --env GROK_API_KEY \
  --placeholder "sk-{rand}" --value "$OPENAI_API_KEY"
sbx run --kit docker.io/ajeetraina777/sbx-grok-kits:openai claude

The cloud tags hold no key. The sbx proxy injects it from the stored secret on the wire, so the key never enters the sandbox (sbx run has no -e flag).

⁠How it works

Each tag installs @vibe-kit/grok-cli via npm, writes /home/agent/.grok/user-settings.json for the right backend, sets GROK_BASE_URL/GROK_MODEL, and (for the cloud tags) adds a narrow serviceDomains/serviceAuth rule so the proxy injects the stored key on outbound requests. Inside the sandbox, run grok for the interactive TUI or grok -p "..." for a headless one-off.

Per-backend setup notes, validation steps, and the raw spec.yaml for each kit live on GitHub: https://github.com/ajeetraina/sbx-kits-grok/tree/main/providers⁠