Ollama struggles in Copilot agents

This is less a setup problem than a capability mismatch: Copilot's agent UX assumes reliable tool calling, context handling, and edit discipline that many local models still do not deliver.

• –VS Code documentation says agent availability depends on model tool-calling support, which explains why "visible in picker" does not mean "good at agent work"
• –Ollama's VS Code integration is becoming easier to wire up, but local coding-agent quality still depends heavily on the chosen model and serving stack
• –The useful path may be high-end local coding models, VS Code Insiders' OpenAI-compatible provider, or dedicated local-first tools instead of expecting Copilot Chat parity
• –The thread captures a real developer gap: local privacy and zero cloud spend are appealing, but cloud Copilot models still set the baseline for agentic editing