Developers use OpenShift Dev Spaces. AI assistants call the MaaS / RHCL AI Gateway (OpenAI-compatible /v1), which forwards to llm-d / vLLM. Source code and inference stay on the cluster when configured that way.
Open the Dev Spaces dashboard in a browser. The default workspace runs OpenCode (custom devspaces-opencode image with Web UI on port 4096). No desktop install required.
Desktop IDEs can attach to the workspace over an authenticated oc tunnel (SSHD on port 2022 is not exposed via Route). See Dev Spaces docs for the SSH flow.
OpenCode is the default AI assistant in che-code workspaces; type: continue / TYPE=continue provides Continue, Cline, and Roo Code. These three extensions are Apache 2.0, speak OpenAI-compatible APIs, and are installable on Open VSX. The charts pre-wire them via ConfigMaps in pca-devspaces when that type is selected.
| Capability | Continue | Cline | Roo Code |
|---|---|---|---|
| Primary UX | Chat, edit, agent, tab autocomplete | Autonomous agent sidebar | Multi-mode agent (Code, Architect, Ask, Debug, …) |
| Tab autocomplete | Yes | No | No |
| File / terminal | Yes | Yes | Yes |
| MCP | Yes | Yes | Yes |
| Config delivery | config.yaml ConfigMap |
Provider ConfigMaps | Provider profiles ConfigMap |
| Self-hosted vLLM | apiBase + model |
OpenAI-compatible base URL | Needs native tool_calls from the model |
Practical split: OpenCode for the default agentic UX; Continue for autocomplete; Cline/Roo for agentic workflows on che-code. See deploy_existing_openshift/README.md and PCA_Deployment_ARO/README.md.
ROSA/ARO default IDE endpoint (MaaS / RHCL front door):
https://maas-default-gateway-data-science-gateway-class.openshift-ingress.svc.cluster.local/v1
Existing OpenShift uses the configured Gateway class (openshift-default in the supplied overlay) or listener hostname. The deployment script prefers the live listener hostname unless maas.hostname is explicitly set.
Continue tab autocomplete uses /local/v1 on that same host (authenticated, skips guardrails and Semantic Router). OpenCode has no second URL.
Each Dev Spaces namespace gets an API key Secret (pca-maas-apikey). Break-glass (no API key): point IDEs at llm-d directly with aiGateway.escapeHatchToLlmd=true.
Agent modes need the model to return OpenAI-style tool_calls. For Qwen3 / Qwen3.6 families, vLLM must set the correct --tool-call-parser and --reasoning-parser (values live in cloud values-*.yaml / LLMInferenceService args). Wrong parsers show up as “model did not call any of the required tools.”
See the ARO deploy guide section on tool calling for the current parser matrix.

