Ollama 0.14-rc2 introduces an experimental agent loop mode that allows LLMs to execute bash commands and perform web searches on your system with interactive user approval. The feature includes safety mechanisms like auto-allowlists for safe commands, denylists for dangerous patterns, and warnings for operations outside the project directory. The release also adds experimental image generation support using the Z-Image model with an MLX-based runner, though performance optimizations are planned for future iterations.
Sort: