Operate the runtime, not the chat.
Millrace is for technical operators, not casual users. It is open source, local, file-backed, and meant to be inspected through CLI commands and workspace files.
Install it, initialize a workspace baseline, compile the graph, run the daemon, then inspect evidence and state instead of trusting conversational memory.
pip install millrace-aiStart local, then inspect every state change.
The default posture is deliberately mechanical: install the package, create the workspace baseline, compile the runtime graph, run one controlled tick or a monitored daemon, and inspect state from the CLI plus files.
pip install millrace-ai
millrace init --workspace "$WORKSPACE" writes the package-managed baseline and local runtime tree.
millrace compile graph --workspace "$WORKSPACE" is the operator-facing graph contract before dispatch.
millrace run daemon --workspace "$WORKSPACE" --monitor basic runs repeated ticks; use millrace run once for a single controlled step.
millrace status, millrace runs show <run_id>, and workspace artifacts expose evidence, closure state, and runtime progress.
Mission Control sits on top of inspectable state.
Use Mission Control as the shell/UI layer for the same runtime posture: CLI commands, compiled graph output, queue files, run artifacts, Arbiter state, and evidence reports remain the source of truth.