Loading...
Loading...
A wave of Git-adjacent projects is reexamining Git as both a storage model and a workflow backbone—while a few high-profile missteps muddy the conversation. Developers are experimenting with alternatives and extensions, from Beagle SCM’s attempt to improve Git’s so-called “blockchain” mechanics to “gitgres,” a prototype that stores Git objects and refs in PostgreSQL to address ecosystem scaling limits. In parallel, Go-based tooling is thriving: Foundry promotes a markdown-in-Git, database-free CMS workflow, and gitgo offers a Go-native Git library. The discourse is amplified by governance topics (Forgejo/PR templates) and a PR embarrassment over a reportedly plagiarized Microsoft Git explainer graphic.
Debates about Git's data model affect how teams store, scale, and collaborate on source and content. Tooling and governance choices influence developer workflows, integration costs, and long-term maintainability.
Dossier last updated: 2026-05-10 18:06:37
A longtime WordPress user migrated their personal site after Bluehost proved unreliable and support inadequate, triggering a reassessment of WordPress’s fit. The author found WordPress good for publishing but ill-suited to treating a multi-decade blog as an inspectable, editable archive. They wanted local workflows, easier cross-linking, intentional taxonomies, and the ability to version and edit content as files. The migration led them to build Yapress, a markdown-first publishing setup (named after their initials) that imports WordPress content, supports taxonomies, series, and content validation, and publishes via npm. The change matters because it reflects a broader shift toward file-based, developer-friendly publishing and ownership over content and tooling.
After two decades on WordPress, the author migrated off shared hosting to Bluehost last November and found the move exposed larger frictions with WordPress for a personal archive site. Bluehost proved cheaper but noisy and unreliable; support and Cloudflare integration didn’t restore confidence. The author wanted a workflow focused on treating years of posts as data—easier local search, reorganization, deliberate cross-linking, taxonomies, and series—rather than a publishing-first CMS. They built Yapress, a markdown-first, file-based publishing tool (using Codex, Claude, and Gemini), with WordPress import, content validation, and npm publication. The shift lets them edit in code editors, use Git, and manage the site as an inspectable archive.
A developer published WUPHF’s new wiki layer: a Karpathy-style, LLM-native knowledge substrate that uses markdown + git as the canonical store, with bleve (BM25) search and SQLite metadata instead of vector or graph DBs. Each agent gets a private notebook and a shared team wiki; notebook entries can be promoted to canonical pages with provenance commits under a dedicated bot identity. The system records append-only JSONL fact logs, rebuilds entity briefs via a synthesis worker, detects broken wikilinks, and runs daily linting for contradictions and staleness. Benchmarked at 85% recall@20 on BM25 over 500 artifacts, it’s MIT-licensed, self-hosted, and integrates with various LLMs; source and demo are on GitHub.
A developer released WUPHF’s open-source LLM-native wiki that stores agent knowledge as markdown + git, using bleve (BM25) and SQLite for indexing instead of vector or graph databases. Agents get private notebooks and a shared team wiki; draft-to-wiki promotion, append-only JSONL fact logs, and a periodic synthesis worker create and update canonical entity briefs. The system emphasizes durability and provenance—commits use a dedicated identity, canonical stable IDs, broken-link detection, and daily linting for contradictions and staleness. Benchmarks show ~85% recall@20 on a 500-artifact test; the project is MIT, self-hosted, and integrates with various LLMs. The author invites deeper discussion on substrate tradeoffs and design choices.