DevSpaces AI (browser IDE)

Run OpenShift DevSpaces with in-browser AI (Che Code + Continue) against this chart’s LiteLLM Route (Granite / Qwen).

Info: Identity

Hub Guest is only for Developer Hub. DevSpaces is started by your OpenShift Sandbox user. The chart prepares Devfiles and .continue/ config; it does not open DevSpaces as Guest.

Prerequisites

  • Chart installed and verified (LiteLLM Route returns models).
  • DevSpaces available on the cluster (true on Developer Sandbox).
  • Quota for ~200m CPU / 1Gi memory request (limits up to ~2 CPU / 3Gi). Stop other heavy workspaces first.

Option A — From Hub template (recommended)

  1. Guest → Create → Agent-friendly DevSpaces AI Workspace.
  2. Choose workspace name, agent language (Python / Node.js / Quarkus), and default Continue model.
  3. The scaffolder registers a catalog Component (managed-devworkspace=true).
  4. Within ~1 minute the agent-applier creates a started DevWorkspace with Continue → LiteLLM prewired (Secret rhdh-agent-sandbox-continue).
  5. Open DevSpaces dashboard and open your workspace — no git push.

Fallback: Component link Language factory URL opens the same language skeleton Devfile via DevSpaces factory.

Option B — From this repository

  1. Use files/devfiles/devfile.yaml or a skeleton under files/templates/skeletons/*/devfile.yaml.
  2. Create a DevWorkspace that clones https://github.com/maximilianopizarro/rhdh-agent-sandbox (or your fork) and uses that Devfile.
  3. Start the workspace; wait until phase Running.

Example (adjust resources if needed):

oc apply -f - <<'EOF'
apiVersion: workspace.devfile.io/v1alpha2
kind: DevWorkspace
metadata:
  name: rhdh-agent-ai-demo
spec:
  routingClass: che
  started: true
  template:
    projects:
      - name: rhdh-agent-sandbox
        git:
          remotes:
            origin: https://github.com/maximilianopizarro/rhdh-agent-sandbox.git
          checkoutFrom:
            revision: main
    components:
      - name: tools
        container:
          image: quay.io/devfile/universal-developer-image:ubi9-latest
          memoryLimit: 3Gi
          memoryRequest: 1Gi
          cpuLimit: "2"
          cpuRequest: 200m
          mountSources: true
EOF

oc get dw rhdh-agent-ai-demo -w

Wire Continue → LiteLLM

The Helm chart creates Secret rhdh-agent-sandbox-continue with:

Key Source
LITELLM_API_BASE LiteLLM Route URL (…/v1)
LITELLM_API_KEY litellm-master-key from rhdh-agent-sandbox-secrets

Inside the workspace terminal, run Devfile command wire-continue. It prefers env vars if set; otherwise it loads that Secret via oc and writes .continue/config.json (models granite / qwen3).

Manual fallback:

export LITELLM_API_BASE="$(oc get secret rhdh-agent-sandbox-continue -o jsonpath='{.data.LITELLM_API_BASE}' | base64 -d)"
export LITELLM_API_KEY="$(oc get secret rhdh-agent-sandbox-continue -o jsonpath='{.data.LITELLM_API_KEY}' | base64 -d)"

Continue is preinstalled on workspace start (DEFAULT_EXTENSIONS + OpenVSX download; CLI fallback). Open the Continue sidebar in Che Code, select Granite (LiteLLM), and send a short chat (for example Reply with exactly: DevSpaces Continue OK). Screenshots of the query → reply and Hub MCP action → reaction loop: DevSpaces journey.

Tip

Use litellm-master-key (via the Continue Secret), not your OpenShift user token. The user token is only for LiteLLM → shared models (model-api-key). Hub tool demos use LiteMaaS via Lightspeed/MCP Chat — not Continue.

What ships in skeletons

Path Purpose
devfile.yaml UDI + modest resources + AI help commands
.continue/config.json OpenAI-compatible models → LiteLLM
.continue/skills/* DevSpaces AI, OpenShift MCP notes, Lightspeed RAG
.vscode/extensions.json Recommends Continue
Devfile DEFAULT_EXTENSIONS + download-continue-extension Preinstalls Continue from OpenVSX (linux-x64 .vsix)
docs/prompts/* Prompts aligned with Hub catalog

AI and MCP from the IDE (two paths)

Continue is one sidebar, two network paths:

Path Flow Secret
AI chat Continue → LiteLLM Route /v1 → Granite/Qwen rhdh-agent-sandbox-continue
Hub MCP Continue → Hub Route /api/mcp-actions/v1 → catalog/TechDocs tools mcp-token (written by wire-continue into mcpServers)

OpenShift/Kubernetes MCP remain ClusterIP and are consumed from Hub Lightspeed / MCP Chat, not from the IDE. Screenshots and prompts: DevSpaces journey.

Info: Continue v2

Continue 2.x stores live keys in ~/.continue/config.yaml (local only). wire-continue also writes placeholder .continue/config.json in the repo tree — do not paste real keys into git.

Stop when done

oc patch dw rhdh-agent-ai-demo --type=merge -p '{"spec":{"started":false}}'

Or use Stop in the DevSpaces UI. Freeing the workspace avoids quota pressure on Hub/LiteLLM.