Local models that don't support tools now just work
When you bring your own local model through an Ollama / OpenAI-compatible endpoint, the assistant now checks whether that model supports tool calling before it uses it. A completion-only or roleplay model (which used to fail the whole chat with a "does not support tools" error) is detected automatically and sent a plain chat request instead — so it works as a chat model rather than breaking the round.
This check runs both when you add a provider in Settings → Language Models and, for models configured on the server, shortly after startup. Tool-capable models are unaffected and keep full tool access.
Running a local model and seeing a context size error instead? Local runtimes load with a small default context window — raise it on the host (for Ollama, set OLLAMA_CONTEXT_LENGTH) so the full prompt fits.