Local providers
Ollama, llama.cpp, and Docker Model Runner connect to local or organization-managed model services. Start the server and confirm the extension can reach its configured address. Local inference can keep model processing on that endpoint; features that use hosted Real Memory services or other network services remain separate.
Hosted providers
Cloud AI providers may require an API key. Obtain credentials from the provider and enter them through the product's credential setting. Do not paste keys into chat or public project files. Provider logging, retention, training, and deletion practices are governed by that provider's terms.
Custom and enterprise endpoints
Use the exact endpoint and authentication policy supplied by your organization. Confirm that the selected model supports the required chat and tool capabilities, and check your organization's rules before sending workspace content.
Readiness and flow selection
The product supports Auto, Direct Model, Single Model, and Planner + Coder flows. The flow determines how the selected model roles are used; it does not change the provider's own data practices. Readiness checks identify missing or unreachable provider settings before a full task begins.
Review the privacy policy and data flow.
Ollama Writer recovery in VS Code
Version 1.6.606 uses the selected Ollama model to propose a different approach in a separate advisory request. The subsequent Writer retry uses the same model and ticket.