Customisation¶

The fully spelled-out docker run commands on each image's page are the reference invocations — everything else here builds on top of them without weakening the hardened defaults. These pages cover the ways to adapt a container's launch, configuration, and model provider to your own setup.

Custom configuration files¶

Each workload image keeps configuration and runtime state in its persistent home volume, and not every configuration file is safe to bind-mount as a read-only input. Custom configuration files explains the difference between input-only files, agent-owned mutable state, and whole configuration directories, and which pattern applies to each case.

Local models¶

Pointing an agent at a local or self-hosted model server is a per-agent, per-server combination that this repository cannot verify exhaustively. Local models is a test plan and a set of candidate configurations — a record of what to test and how, not a set of proven recipes — for routing an agent's traffic to llama.cpp, Ollama, vLLM, LM Studio, or another local server instead of a hosted provider.

Shell shortcuts¶

The reference docker run commands are deliberately explicit about every containment flag, which makes them tedious to retype. Shell shortcuts shows how to wrap each one in a shell function that mounts the current directory and starts the matching image, while keeping the image-specific flags in one auditable place.