opencode

To configure a multi-provider terminal coding agent, install the OpenCode CLI, log in to your providers, and define model routing in opencode.json using the <provider>/<model-id> format. You handle secrets securely via {env:VAR} interpolation.

The best way to route coding tasks is to define plan and build agents in opencode.json, assigning read-only models to the plan agent and writing models to the build agent. You specify models using <provider>/<model-id> prefixes for accurate routing.

To run terminal coding workflows in CI, use the headless command `opencode run --json` to output JSONL logs. You can also use `opencode serve` or ACP to integrate external orchestration for automated review pipelines.

Yes, you can configure MCP servers within opencode.json to enable external orchestration. This allows the terminal coding agent to interact with MCP/ACP integrations for automated tool execution and workflow management.

To securely manage API keys for a CLI coding agent, use {env:VAR} interpolation in your opencode.json configuration file. This ensures your provider secrets are loaded from environment variables rather than being hardcoded.

Yes, opencode supports custom agents with restricted tool access. You can define custom agents in opencode.json and control their tool allowlists to create safer review workflows and limit specific agent capabilities.