Skip to content

Guided llama.cpp install mission in KubeStellar Console #21973

@clubanderson

Description

@clubanderson

We built a guided install mission for llama.cpp inside KubeStellar Console, a standalone Kubernetes dashboard (unrelated to legacy kubestellar/kubestellar, kubeflex, or OCM — zero shared code).

Open the llama.cpp install mission

What the mission does

The mission deploys llama-server as a Kubernetes workload with a PVC for the GGUF model cache, and exposes it via a ClusterIP Service on the OpenAI-compatible /v1/chat/completions endpoint. Each step:

  1. Pre-flight — checks CPU vs CUDA image selection, verifies NVIDIA device plugin when GPU is requested
  2. Commands — deploys the ghcr.io/ggml-org/llama.cpp:server image with an initContainer that fetches a GGUF model from Hugging Face on first start
  3. Validation — waits for rollout and health probe, then curls /v1/chat/completions to confirm inference works end-to-end
  4. Troubleshooting — on failure, reads pod logs and suggests fixes (initContainer download failures, OOM, readiness probe timing, CPU vs GPU image mismatch)
  5. Rollback — complete uninstall path for the Deployment, Service, PVC, and namespace

The mission also ties llama.cpp into the Console's agent selector: set LLAMACPP_URL to the in-cluster Service URL and llama.cpp appears as a chat-capable provider in the dropdown, so Console chat routes through your local llama-server.

Why we're reaching out

The Console now ships with six local-LLM runner integrations (Ollama, llama.cpp, LocalAI, vLLM, LM Studio, Red Hat AI Inference Server) so operators in regulated or air-gapped environments can keep LLM traffic inside their cluster. llama.cpp is the "dependency-minimal" option — easiest to audit, lightest to run, supports the broadest hardware matrix.

Install

Local (connects to your current kubeconfig context):

curl -sSL https://raw.githubusercontent.com/kubestellar/console/main/start.sh | bash

Deploy into a cluster:

curl -sSL https://raw.githubusercontent.com/kubestellar/console/main/deploy.sh | bash

Mission definitions are open source — PRs welcome at install-llama-cpp.json. Feel free to close if not relevant.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions