Version v2.12.0

Setup wizard

opencode-rag setup is now the single entry point and asks for the basics:

• provider (Ollama, OpenAI-compatible, Cohere, Anthropic, Google), base URL and API key
• embedding model (required), description model, vision model and decision model (tev1)
• optional: reranking, shell guard, quirk capture, ClickHouse backend

Model lists are probed from the host (/api/tags + /api/show for Ollama, /v1/models elsewhere), and the embedding dimension is probed and stored so the vector-store schema stays stable.

Provider/model settings are written to the global config (~/.config/opencode/opencode-rag.json); feature toggles stay in .opencode/opencode-rag.json. Stale provider keys left in a workspace by the old template are detected and removed (with a .bak backup).

Breaking changes

• opencode-rag init was removed. setup scaffolds the workspace and re-syncs it on every run.
• Re-open the wizard with setup --reconfigure; script setups with --yes plus --provider, --base-url, --api-key, --embedding-model, --description/--vision/--decision [on|off], --vector-store, --skip-runtime, etc.

Also in this release

• TUI: new /rag-settings slash command.
• Docs updated across cli, installation, configuration, development, plugin, troubleshooting.