OpenCode & Coding Agents
OpenCode is the reference runtime — not the platform contract.
The first Developer Workspace profile uses OpenCode as an AI coding agent running within a reviewed Kubernetes Namespace-based execution boundary.
What the workspace provides
- A source checkout associated with a declared repository and revision.
- A replaceable coding-agent runtime, with the workspace filesystem and constrained compute resources.
- An explicitly selected inference service, rather than unrestricted access to every model endpoint on the network.
- Scoped Git, network and tool/MCP capabilities.
- A declared verification command, with independent checking of the agent's reported result.
- Evidence export and controlled cleanup for ephemeral workspaces.
These are candidate contract semantics under validation, not guarantees of a generally available hosted coding workspace.
OpenCode and the second-runtime proof
The ADR-Platform-039 live-proof package documents a bounded trial in which OpenCode and Codex ran against the same reviewed workspace profile with a declared source fixture and verification steps.
Both runtimes were exercised under the proof's explicit constraints, supporting runtime replaceability as an architectural direction. This does not prove that every coding agent, model or workload is supported.
Relationship to OK AI
OK AI can provide model-serving and agent-related infrastructure. Developer Workspace is different: it controls the coding execution environment, source access, filesystem, credentials and lifecycle.
A coding workspace may consume an approved inference endpoint as a capability. It must never gain infrastructure authority merely because that endpoint or the workspace runs inside OpenKubes.
Current scope
There is no turnkey workflow for production use. For evaluation on a disposable cluster, the reference implementation provides make live-up and make live-down, which bring one OpenCode workspace up and remove exactly what they created. Broader access control (a requester role) and production lifecycle operations remain implementation work.
To follow a guided evaluation, see Your First AI Coding Workspace.
See the validation status before evaluating the proof for your own environment.