# Decisions

## 2026-07-07 - Build A Go CLI First

Decision:

Implement `gemcms` first as a Go command-line tool.

Rationale:

- A single static binary fits the self-hosted Gemini audience.
- The core workflows are filesystem and text processing tasks.
- Avoiding a web admin panel keeps the first version simple and robust.

Consequences:

- The first user experience is terminal-first.
- A TUI or local web editor can be added later without changing the capsule format.

## 2026-07-07 - Keep Capsules File-Based

Decision:

Store pages, articles, menus, widgets, and config as project files. Do not use a database.

Rationale:

- Gemini capsules are naturally static and text-first.
- File storage keeps content easy to edit, back up, diff, and publish.
- A database would add operational cost before it solves a real problem.

Consequences:

- Build-time validation and generation are important.
- Future features should preserve direct file editability.

## 2026-07-07 - Treat Widgets As Build-Time Blocks

Decision:

Widgets are static Gemtext blocks expanded during `gemcms build`.

Rationale:

- Gemini has no browser-side runtime.
- Static blocks keep output simple and portable.
- Menus, recent articles, indexes, and similar blocks can be generated reliably.

Consequences:

- Widget behavior is deterministic at build time.
- Interactive runtime widgets are out of scope for the core CMS.

## 2026-07-07 - Import Existing Content As A First-Class Workflow

Decision:

Support importing one file or a directory of Markdown/Gemtext files through `gemcms add`, and support `gemcms create capsule` for creating a capsule, importing content, and optionally building output in one command.

Rationale:

- Many users will already have notes or articles as `.md` or `.gmi` files.
- A CMS for Gemini should reduce setup friction rather than requiring manual folder work.
- Keeping imports deterministic and file-based preserves direct editability.

Consequences:

- Import behavior must infer metadata from frontmatter, headings, and filenames.
- Directory imports should skip unrelated files rather than fail on them.
- Health/status commands become important for confidence after bulk imports.