Wingman-AI setup: providers, MCP tools and code indexing

Uploaded March 21, 2025, this walkthrough covers Wingman-AI as a VS Code extension. The speaker describes it as an open source coding assistant with continued Ollama support for a local LLM workflow, alongside hosted providers. No Wingman-AI version number or separate recording date is supplied. The settings and capabilities describe that upload's context.

The setup separates the main AI provider from the embeddings provider used for code indexing. Anthropic is the demonstrated default, with an API key required for that configuration. The speaker uses Claude 3.7 Sonnet and discusses 3.5 as an option, plus thinking and raw modes for 3.7. He recommends a smaller autocomplete model and explains controls for surrounding file context and output length. He says the extension does not provide tab completion in this setup.

Agent settings include vibe mode, which automatically accepts file edits and commands. The speaker says Wingman blocks destructive commands and lets users undo edits; he does not test those safeguards in this settings tour. With vibe mode off, users approve or reject actions individually. Experimental autofix sends errors found after edits back to the model for another attempt.

The tour also explains MCP tools launched through commands or connected through SSE, with configuration saved per workspace. Code indexing uses generated summaries and file inclusion rules. A project-root Wingman rule file supplies extra instructions. The video ends before a project-generation demonstration, which the speaker reserves for a follow-up.