Local-first AI development without losing the thread
Keep repositories and development tools where they already live while preserving durable product context.
Developers already have a working environment: repositories, shells, Git configuration, SSH access, package managers, test tools and local conventions. An AI workflow should respect that environment.
Local is a boundary, not a slogan
“Local-first” should lead to practical answers. Where does the application run? Which data stays on the device? What can synchronize? What does an agent provider receive? Are terminal transcripts captured? The answers must follow the shipping implementation.
Context still needs structure
Keeping execution local does not automatically preserve why a change exists. Requirements, decisions, agent runs, commits and verification evidence can still become disconnected. A dependable local workflow needs both environment compatibility and a persistent product thread.
Workbench is being designed around that combination: real local development tools, plus structured relationships from intent to evidence.