Recently updated open-source projects

iContainer is a native SwiftUI app that puts a graphical layer over Apple's official container CLI. It manages containers, images, lifecycle actions, logs, resource statistics, and interactive shells, and can optionally explain logs with Apple's on-device Foundation Models. The app requires macOS 26 or later on Apple silicon and an installed Apple Container CLI. Apple Intelligence is needed only for the local AI log analysis; the remaining management features work without it.

Dockge manages compose.yaml-based stacks through a web interface for creating, editing, starting, stopping, restarting, and updating services, with live progress and terminal output. Compose files remain in a normal directory structure and can still be used with standard Docker Compose commands, while agents can connect multiple hosts. It targets Linux stack management and is not intended for standalone containers without compose.yaml.

visual-style-ppt turns documents, articles, outlines, topics, or visual references into a style-led presentation workflow. It extracts or selects a reusable visual direction, plans the slide sequence, writes page-specific prompts, and generates each slide as a polished Image 2 image; those images can then be assembled into an image-only PPTX. The result prioritizes visual finish over editable PowerPoint layouts. Image 2 is mandatory for every generated page, and PPTX packaging also requires a compatible assembly environment.

Future AGI is a self-hostable open-source platform for observing and improving LLM or agent applications. It combines tracing, evaluations, simulations, datasets, guardrails, an AI gateway, and prompt optimization in one feedback loop. Traces use OpenTelemetry, while the gateway exposes OpenAI-compatible HTTP and integrates with many providers and frameworks. A full deployment includes PostgreSQL, ClickHouse, Redis, RabbitMQ, and Temporal, so it best fits teams that need unified quality and observability infrastructure.

OpenMemory adds persistent memory to LLM applications through Python and Node SDKs or a multi-user backend with an HTTP API, dashboard, and MCP endpoint. It models different memory sectors, temporal relationships, decay, and reinforcement rather than relying only on similarity search, and exposes traces showing why a memory was recalled. Storage can remain self-hosted in SQLite or PostgreSQL, making it useful for agents and copilots that need durable memory under their own control.

TestMu AI Agent Skills add framework-specific testing guidance to coding assistants such as Claude Code, Copilot, Cursor, and Gemini CLI. The repository covers major automation stacks including Selenium, Appium, Playwright, and Cypress across numerous languages, with skills installable individually. Using the hosted testing platform requires Node.js plus a TestMu AI account, username, and access key.

Codex++ patches the locally installed Codex Desktop app so a small runtime can load tweaks from a user-owned folder. Those tweaks may alter the interface, add settings pages, run main-process code, or use OS-level features through its native bridge, while runtime files, configuration, logs, backups, and tweak data remain outside the app bundle. A watcher can reapply the patch after Codex updates. This is an unofficial modification that executes local extension code, so users should rely on its diagnostics, safe mode, repair, and uninstall controls.

jiamu-skills bundles 11 workflows for local AI agents, covering problem framing, root-cause analysis, multi-angle decisions, sales content, China A-share research, magazine layouts, TV commercials, video downloading, and GIF splitting. Once installed, the relevant skills can be triggered through natural-language requests. Reasoning-oriented skills can also serve as prompt guidance, while utility skills require file and shell access.

SDAI gives Android and iOS users one mobile interface for generating images through a self-hosted AUTOMATIC1111 or SwarmUI server, several hosted image APIs, crowdsourced workers, or supported on-device diffusion runtimes. Users can select models and generation controls, keep results in a local gallery, inspect metadata, share images, and export the gallery as a zip file. Capabilities vary by operating system, app flavor, and provider. Remote backends generally require a reachable server or API key, while local generation depends on compatible model files and sufficient device hardware.

Mattermost Desktop is an Electron client for Windows, macOS, and Linux that keeps access to Mattermost servers in one desktop workspace. It supports a server dropdown, dedicated tabs for channels, boards, and playbooks, desktop notifications, unread badges, deep links, and background operation. The client connects to an existing Mattermost server rather than providing a hosted chat backend, so a server URL is required at first launch.

Kavita turns manga, webtoons, comics, EPUB files, and PDFs into a self-hosted reading library with responsive web readers for phones, tablets, and desktops. It includes metadata search, collections, reading lists, and multi-user controls with roles and age restrictions. The project is still labeled beta until version 1.0, so users should keep backups in case behavior or stored data changes.

This Claude Skill shapes conference talks, seminars, thesis defenses, and grant briefings around a clear argument rather than a collection of topic slides. It uses takeaway-style titles, keeps results slides focused on one exhibit, and applies in-slide citations plus a final references section. Claude's built-in PPTX capability handles the actual file generation, so code execution and file creation must be enabled.

Seedance2-skill turns a skill-aware agent into a video concept workspace for Seedance. It can work from text, images, video, or audio, branch into several ideas when given only an image, and evaluate prompts for memorability, surprise, emotional movement, and narrative change. A reference library supplies more than one hundred cinematography terms and style cues. Once a concept is ready, the included Python CLI can submit generation requests across Seedance 2.0, 1.5 Pro, and 1.0 model families.

Reticulum MeshChat discovers peers and exchanges encrypted messages, images, files, voice recordings, and low-bandwidth audio calls across local Ethernet or Wi-Fi, internet relays, LoRa radios, and other Reticulum interfaces. New users create an identity, choose a display name, announce themselves, and configure the network links they need, then use either the web interface or a standalone application. Audio calling remains a beta feature and, when using the command-line web interface, generally requires a secure browser context and works best in Chromium-based browsers.

WebTorrent Desktop packages torrent streaming into a graphical app for macOS, Windows, and Linux, with installers available from the official site or release page. It is designed to start playing media before the full download completes. The app reports basic usage statistics for product improvement but states that it does not send identifying information or track which torrents are added.

CoWork OS is a GUI-first desktop application with a companion `cowork` CLI for running AI-assisted work in one local workspace. It brings coding, email, research, browser tasks, documents, spreadsheets, presentations, agent sessions, and automations into a shared operator console instead of separate apps. Long-running tasks can be organized through session timelines, boards, teams, plugins, and multiple model providers, while normal local CLI use shares the desktop configuration. It is aimed at users who want a broad, locally governed environment for complex AI workflows.

Tabby offers a self-hosted alternative to cloud coding assistants and connects to developers through IDE extensions. Its OpenAPI interface can be integrated with existing infrastructure, while the server is self-contained and does not require a separate database or cloud service. It also supports consumer-grade GPUs for teams that want local control over models and code context.

Install Hono Skill when you want an AI coding agent to work with Hono using an inline reference instead of guessing framework APIs. It covers routing, context, middleware, JSX, validation, RPC, streaming, helpers, and request testing through the Hono CLI. The skill can be added through the Claude Code marketplace or a cross-agent skills installer. An optional hono-docs MCP server is available when the agent also needs to search and retrieve the latest Hono documentation.