GitHub Trending on 14 August: diagram-design leads, plus four developer freebies (2026)

GitHub's 14 August daily board is unusually concentrated: diagram-design added 4,475 stars, while macro added 1,239. The useful split is between tools that change how you build — agent skills, local inference, and routing — and free developer offers with a clear action attached. 1

The fastest climbs

The table is sorted by stars gained in the daily window, rather than by the board's display order. Total stars and primary language are the values shown on GitHub's daily page; the final column adds the README-level detail needed to decide whether to open the repo.

Repository	Stars today	Total stars	Language	What it is and why it matters
cathrynlavery/diagram-design	+4,475 1	15,134 1	HTML	An agent skill for Claude Code, Codex, and Pi with 27 diagram types. The current README adds semantic patterns and optional accessible motion while keeping static HTML as the default. 2
macro-inc/macro	+1,239 1	2,680 1	Rust	An all-in-one workspace for email, messages, docs, tasks, agents, and CRM. Its README says the product is built in SolidJS and Rust and is fully open source under AGPLv3, with self-hosting available. 3
msitarzewski/agency-agents	+778 1	145,265 1	Shell	A large collection of specialist agent files with defined workflows and deliverables. A native app can install the roster into tools such as Claude Code, Cursor, Codex, and Gemini. 4
cactus-compute/needle	+769 1	5,057 1	Python	A 45-million-parameter tool-calling and structured-extraction model packaged as a 14 MB binary. The README says a full session uses about 28 MB of RAM and inference runs without a network after the engine is fetched. 5
semantica-agi/semantica	+713 1	6,869 1	Python	A self-hostable infrastructure layer that builds context and knowledge graphs for agents, with provenance and decision history. The current README recommends the v0.6.5 security update and flags a simple rule matcher that needs validation before production compliance work. 6
NVIDIA-NeMo/Switchyard	+408 1	1,279 1	Rust	A proxy and library for LLM traffic that translates OpenAI and Anthropic formats, routes across backends, and records Prometheus metrics. The project calls itself pre-alpha and says it is not for production use. 7
unslothai/unsloth	+328 1	71,142 1	Python	A desktop and web UI for running, training, and deploying local models, with integrations for Claude Code, Codex, MCP, and free training notebooks. 8
anthropics/skills	+312 1	169,134 1	Python	Anthropic's public examples of Agent Skills: self-contained folders of instructions, scripts, and resources, including document, PDF, spreadsheet, and MCP-building examples. 9

Three worth opening first

Diagram Design is a follow-up, not a rerun. It was already on yesterday's board, but the current README has a real change to inspect: semantic system patterns now separate behavior from layout, and motion is optional with a static fallback. If your bottleneck is making architecture diagrams, flows, or timelines that look like part of a real product, install it in a test project and compare its output with your current agent workflow. 2

Macro is the largest product-shaped climb. Its pitch is not another isolated chat or task app. Email, messages, docs, tasks, CRM, pull requests, and agents share a searchable workspace with bidirectional links. The project has a hosted app, but the repository also documents local operation and AGPLv3 self-hosting. That makes it worth opening if you are evaluating an all-in-one internal tool; it also means you should read the license and deployment requirements before treating it as a drop-in SaaS replacement. 3

Needle is the cleanest hardware constraint story. A 14 MB model, roughly 28 MB of session memory, structured JSON output, confidence scores, and no network inference after download put it in a different category from cloud agent demos. It is a sensible experiment for an edge device or a local tool router. Its README positions it among other small models, so do not read today's star count as evidence that it matches a frontier model. 5

Free developer value to claim or apply for

1. OpenAI Codex for OSS — apply if you maintain a public project

OpenAI's official program page says core maintainers and people running widely used public projects can apply. The package includes six months of ChatGPT Pro with Codex, conditional access to Codex Security, and API credits through the Codex Open Source Fund for pull-request review, maintainer automation, release workflows, and other core open-source work. If your project matters to the ecosystem but does not fit the obvious criteria, the page says to explain that role in the application. 10

Action: Apply through the official Codex for OSS form. The program page says "Apply today" and links to that form. 10

2. GitHub Student Developer Pack — stack tools while you are enrolled

The Pack bundles GitHub and partner offers for students. The current page lists GitHub Pro, Codespaces, GitHub Pages, JetBrains, MongoDB Atlas credits, and Heroku credit of $13 per month for 24 months, among other tools. Some partner offers have their own rules; for example, the page lists different Azure offers for students aged 18+ and 13–17. 11

Action: Sign up for the Student Developer Pack, then check each partner's terms before you build a project around a benefit. GitHub's page also currently says new GitHub Copilot plan sign-ups are temporarily paused, so do not count Copilot as an immediately available Pack benefit. 11

3. Gemini API free tier — start without a billing account

Google's billing documentation says new accounts begin on the Free Tier, which provides access to certain Gemini API models within their free rate limits. You do not need a billing account or card for that tier. Higher rate limits, advanced models, and the setting that keeps prompts and responses from being used to improve Google products require a Paid Tier; Google says moving there involves linking billing and prepaying at least $10. 12

Action: Check the current Gemini API billing rules before creating a production key. Free tier is the low-friction route for a prototype; paid billing is a separate decision with a balance requirement.

4. OpenRouter's free model pool — test small models without a card

The current OpenRouter page lists Liquid AI's LFM2.5-2.6B at a free price with a 128K context window. The model page says it fits agent workflows, extraction, and RAG, while Liquid advises against agentic coding or knowledge-heavy tasks. It also warns that prompts and outputs may be retained and used to train Liquid models. 13

OpenRouter's current free-tier guidance says models with the :free suffix get 50 requests per day and 20 requests per minute with no credit card. The limit rises to 1,000 requests per day with at least $10 in credits, and the free pool changes as models come and go. 14

Action: Use the model page for a small extraction or RAG test, and treat the retention note as part of the eligibility check. Do not send sensitive project data simply because the price is zero.

What to star, test, and claim

Star or test diagram-design if your recurring pain is explaining systems visually. The new semantic-pattern and motion work is the reason to revisit it today. 2
Test needle when RAM, bandwidth, or offline operation matters more than frontier-model breadth. Its README gives you a concrete install path: pip install cactus-needle. 5
Inspect macro if your team is tired of context split across mail, chat, docs, and tasks. Read the AGPLv3 and self-hosting material before deployment. 3
Use a caution label on Switchyard and Semantica. Switchyard is explicitly pre-alpha and not for production; Semantica's README calls for a security update and says its current rule matcher needs validation against your own rules. 6 7
Claim the offer that matches your status: maintainers should apply for Codex for OSS, enrolled students should check the Pack, prototype builders can start on Gemini's free tier, and small-model testers can try OpenRouter's free pool. The useful freebie is the one whose eligibility and data trade-offs fit the project you actually have.