Why a local AI hub?
The easy version of an AI hub is a gallery of cards. The difficult version is a place where every card still points to something a team can own: a repository, a revision, a runnable application, a discussion, and a history.
OpenFace starts with that second definition.
The durable objects already exist
Git already knows how to preserve files, revisions, branches, tags, authors, and review. Docker already knows how to package a web application and its runtime. OpenFace does not hide either one. It uses Forgejo as the durable collaboration layer, then adds discovery and local execution around it.
That choice changes the meaning of the interface. A model card is not a decorative record in a separate database. A Space is not an iframe pointed at an unrelated demo. A Prompt revision is not overwritten when a new draft appears. The visible catalog remains connected to repository truth.
Local-first is an operational boundary
“Local” does not merely describe localhost. It describes who controls the failure modes.
- Repository data stays in the Forgejo volume you operate.
- CPU-capable Spaces build and run on the Docker host you selected.
- TLS terminates at the shared gateway.
- Pages are produced from repository branches or directories.
- Agents act through scoped accounts and leave visible issue and pull-request history.
The result can still be shared over a LAN or Tailscale. Local-first means that remote access is an explicit route into your system, not a migration of ownership out of it.
The portal is a lens, not a second source of truth
OpenFace's Next.js frontend makes repositories easier to browse as Models, Datasets, Spaces, Skills, MCPs, and Prompts. The portal may cache or reshape data for presentation, but Forgejo remains authoritative for source, permissions, history, and collaboration.
This avoids the most expensive form of drift: a beautiful catalog that no longer matches what can be cloned, reviewed, or run.
What you trade for that control
Ownership has a cost. The operator must understand Docker resource limits, backups, repository permissions, and the security implications of building third-party Dockerfiles. OpenFace makes those boundaries inspectable; it cannot make them disappear.
That is why the project includes an operations guide, visual QA, and an independent agent review gate. A local platform earns trust through evidence, not through the absence of warnings.