From mixed-state repo to V1
The old repo had three conflicting stories:
- a Python MCP service,
- an unfinished multi-runtime “Universal AI Coding Agent” product,
- a stock nested Turborepo starter.
Documint V1 resolves that by centering one real product path:
- Next.js for the public site and dashboard,
- Python for backend orchestration and MCP,
- Git as the source of truth,
- Documint documenting itself first.
What changed in the control plane
The original self-dogfood backend was process-local. The current platform now persists:
- workspaces and workspace membership,
- connected projects and GitHub source records,
- repo config and artifact definitions,
- verification runs and open findings,
- patch drafts and PR records,
- publish deployments and hosted public pages,
- activity events and CLI tokens.
The result is that /app, the hosted CLI, and /mcp now all read the same stored project state.
Explicit non-goals for the first public deployment
- no full visual CMS,
- no silent autonomous pushes to default branches,
- no enterprise-first compliance program,
- no promise that every historical Rust, Go, or VS Code experiment is part of the launch path,
- no requirement that Hugging Face be online for drift detection, publish, or traceability to work.