Overview

This week's open-source momentum centers on AI-agent infrastructure: context and memory layers (OpenViking, semantica-agi, ai-memory), agent runtimes and workspaces (freebuff, elizaOS), and LLM traffic routing (NVIDIA Switchyard). Local and on-device AI also drew heavy interest, from Needle's 14MB embedded model to Unsloth's desktop trainer and Apple Silicon tooling like oMLX and MLX LM. One caveat: the weekly feed did not populate per-repo star-gain figures for this period, so exact deltas aren't shown below; the leaderboard preserves the feed's pre-sorted order, and totals are included only where the data provides them.

The Leaderboard

No weekly star deltas were present in this week's feed; entries follow the feed's pre-sorted momentum order.

1. cathrynlavery/diagram-design

A collection of 38 editorial diagram types delivered as self-contained HTML + SVG, aimed at AI coding agents such as Claude Code, Codex, and Pi. Version 2.0 adds the Loop flywheel pattern with a shared-memory hub, and 2.3 adds semantic system patterns and optional accessible motion.

2. volcengine/OpenViking

An open-source context database for AI agents that unifies agent memory, knowledge RAG, and skills. It stores memories and resources as queryable context, with docs, a live demo, and community channels linked from the repo.

3. basecamp/omarchy

An opinionated Linux distribution by DHH, positioned as beautiful and modern. The repo hosts the authoritative manual (mirrored to learn.omacom.io), including getting-started guides for users coming from Mac or Windows.

4. cactus-compute/needle

Needle 2 is an open 45M-parameter model for tool calling, device use, and structured extraction, compressed into a single 14MB binary (CQ2-bit) that runs a full session in roughly 28MB of RAM. It targets phones, wearables, smart home devices, and robots.

5. semantica-agi/semantica

Graph-native infrastructure for context and accountable AI systems: it ingests enterprise data, builds a context/knowledge graph, and layers graph analytics and causal reasoning on top with full decision provenance. The project describes itself as an open-source Palantir for AI agents.

6. NVIDIA-NeMo/Switchyard

A Rust proxy and library for LLM traffic that routes requests across models and providers while preserving native OpenAI and Anthropic API compatibility. It translates between API formats and records operational metrics for benchmarking and cost/performance optimization.

7. modular/modular

The repository for open-source components of the Modular Platform, including the MAX framework for model serving and the Mojo language. It positions Modular as a unified platform for AI development and deployment.

8. harry0703/MoneyPrinterTurbo

A one-stop AI short-video generator: given a topic or keyword, it automatically produces a script, matches stock footage, generates subtitles and background music, and composites an HD video via an automated workflow. The project ships a WebUI and API.

9. public-apis/public-apis

The long-running, community-maintained collective list of free APIs for developers and prototyping, continuously updated across hundreds of categories.

10. megadose/holehe

An OSINT tool that checks whether an email address is registered on sites like Twitter and Instagram by probing forgotten-password flows. It returns per-site registration status and associated metadata.

11. AprilNEA/OpenLogi

A native, local-first alternative to Logitech Options+ written in Rust: remap buttons, DPI, and SmartShift over HID++ with no account and no telemetry. It's under active development and not yet stable.

12. lightningpixel/modly

A desktop app for Windows, Linux, and Apple Silicon that generates 3D meshes from images or text prompts using open-source AI models running entirely on the local GPU.

13. macro-inc/macro

An all-in-one team workspace that unifies email, chat, docs, tasks, agents, calls, and CRM in a single interface, with @-linked items and shared team-level AI memory.

14. unslothai/unsloth

Unsloth's first desktop app for running and training LLMs and diffusion models locally. It covers models including Qwen3.8, Kimi K3, MiniMax-H3, Gemma 4, DeepSeek-V4, and FLUX.

15. akitaonrails/ai-memory

A long-term memory solution for AI coding CLIs that lets you switch mid-task between agents such as Claude Code and OpenAI Codex in the same directory without re-explaining architecture or open questions. Linux is the primary supported target.

Also Climbing

Sources