Stacks and inference

Stack composition

Pick one option per layer in the wizard’s Stack step, or run tevrin stack catalog. The options come from a versioned catalog of frontend, backend, data, version control and deployment hosts, adapted from the stack-packs. An Owner can change the stack until the implementation plan is locked, and again when starting a version.

  • The catalog is data under backend/src/Aide.Infrastructure/Agents/Stack/ (catalog.json, packs/, agents/, core/). It is content-hashed.
  • The backend owns the compatibility rules. POST /api/stack/options and /validate serve the wizard, Settings, the API and the CLI.
  • Each release version pins the composition and the catalog version it was built with. Each run records the packs used.
  • Generic specialist agents load the skills of the chosen packs. Spec sections, plan rules, protected paths and the review checklist come from the composition.
  • Projects that used the v2 presets were converted to equivalent compositions.

This catalog describes the stacks of the projects Tevrin builds. It is separate from Tevrin’s own stack.

The origin model

The origin model says how the browser reaches the backend: same-origin, proxy or cross-origin. The options depend on the chosen frontend, backend and hosts. Changing the frontend, backend or their hosts resets it, with a warning.

Stack inference and the Analysis phase

For a repository that already has code, the wizard can infer the stack instead of asking you to choose.

  1. The version starts in an Analysis phase before Refinement.
  2. A Stack Analyst agent reads the repository and writes a stack_profile.
  3. An Owner fills the blanks, edits it and approves it. This is required in every approval mode.
  4. Hosts and deployment setup live on the Analysis page.

An approval is refused while layers are unresolved (STACK_PROFILE_INCOMPLETE). A repository with no code can’t be inferred (STACK_INFER_NO_CODE). From the command line, use tevrin stack detect and tevrin analysis, see the CLI guide.

Database migrations for generated projects

  • Tevrin itself runs only Supabase BaaS migrations, just before the merge.
  • For ORM stacks, the platform runs the data pack’s migrate command: pre-deploy, at container start, or in the production build when AIDE_RUN_MIGRATIONS=true.