llama.cpp puts seven agent tools behind Docker

llama.cpp can now run 7 built-in agent tools inside a reusable Docker isolate rather than directly on the host, covering file reads, glob and grep search, shell execution, writes, edits, and time lookup. The merged Docker-runtime change adds two runtime modes—create a container that is cleaned up with the server, or reuse an existing container—and still labels the feature experimental. For builders extending the case for terminal-native agent tooling, the useful shift is containment, not autonomy: local execution becomes easier to test without making the workstation itself the default sandbox.