git-forums Turns GitHub Discussions Into a Full Forum Site With No Backend

·1 views

A developer has released git-forums, an open-source forum template built with SvelteKit that uses GitHub Discussions as its entire backend, eliminating the need for a separate server or database. The project renders a repository's Discussions section as a fully functional forum website, with categories becoming forum sections and individual discussions appearing as threaded posts. The static site is generated using Svelte 5, SvelteKit, and Tailwind CSS v4, and can be deployed directly to GitHub Pages via GitHub Actions. All core forum features — including user authentication, moderation, reactions, search, and markdown support — are handled natively by GitHub, requiring no additional infrastructure. The template is aimed at open-source projects, developer communities, and teams already on GitHub that want a lightweight forum without managing extra platforms.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

AI Cost Controls Are Shipping With the Meter, Not Years Later

Databricks launched its Genie AI assistant on a pay-as-you-go model this month, bundling budgets, alerts, and per-user spending limits at the same time as the metering feature. The move reflects a lesson the cloud industry took roughly a decade to learn: metering without governance drives churn rather than revenue. Cloud cost management, known as FinOps, only matured as a formal discipline between 2015 and 2019, years after runaway bills had already surprised finance teams. AI platforms are compressing that same adoption-to-governance arc into quarters rather than years. Analysts and educators warn that companies should calculate unit economics — cost per query, per agent run, and per answered question — before an unexpected invoice forces the conversation.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

OpenAI Codex Desktop App Lets Users Create Custom Animated Coding Pets

OpenAI's Codex Desktop App now includes a feature that allows users to generate personalized animated pets to accompany their coding sessions. To get started, users must install a skill via the command '$skill-installer hatch-pet' within the Codex chat interface, then reload skills through the app's keyboard shortcut. Once activated, users can describe any creature or character in a text prompt, and Codex automatically generates animated sprites, a sprite atlas, and a full pet package without requiring any design skills. The custom pet can be selected under Settings and will visually react to Codex activities such as thinking, processing, idling, or encountering errors. The feature is exclusive to the desktop application and is not available on the web version of Codex.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer builds Mandarin learning platform with 100k+ TV clips using Rust and Next.js

A developer has launched MandarinClips, a language learning platform that indexes over 100,000 dialogue segments from Chinese movies and TV series to support immersive learning. The platform is built with a Rust backend using the Axum framework and a Next.js 15 frontend, running on a single two-core VPS. To handle fast search across a large dataset, the project uses PostgreSQL with the pg_trgm trigram extension and GIN indexing, enabling fuzzy text searches in under one millisecond. Media storage is handled via Backblaze B2 proxied through Cloudflare CDN, eliminating egress costs. The project was shared on DEV Community as a technical deep dive into the architecture and engineering decisions behind the platform.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Design Human-in-the-Loop AI Systems That Balance Safety and Scale

Fully autonomous AI systems tend to make costly errors, while fully manual review pipelines are economically unviable, so production AI products must find a middle ground. Engineers determine where human oversight is needed by weighing three factors: the cost of a potential error, whether an action can be reversed, and the model's confidence score on a given decision. Confidence thresholds — numeric cutoffs below which a model's output is routed to a human reviewer — must be calibrated per task type, ranging from 0.7 for support ticket classification to 0.99 for financial decisions. Poorly calibrated thresholds, such as a model that outputs high confidence regardless of accuracy, can be more harmful than having no threshold at all. Architectural patterns like human-approval gates before irreversible actions help teams capture dangerous decisions without reducing the AI system to a costly manual workflow.

0 comments Read more at DEV Community