Open Source Radar — October 13: free tokens, 1-bit models
Today's trending page splits two ways: projects that make frontier coding agents cheaper to run, and projects that make models tiny enough to run anywhere. Both are the same underlying play — pushing the cost of agentic coding toward zero.
free-claude-code (Python, ~57,000 stars) is an independent open-source router that lines up 59 free-tier model providers and lets you point eleven coding-agent harnesses — Claude Code, Codex, OpenCode, Aider and the rest — at them from one catalog. It's on today's daily trending with roughly 57,000 stars and a simple pitch: over a billion free tokens a month, plus automatic failover to your next configured model when a provider goes down mid-turn. Use it as the cheap safety net under a paid coding setup; note it's explicitly not affiliated with Anthropic, and provider terms control how long the free tiers last.
BitNet (C++, ~41,000 stars) is Microsoft's official inference framework for 1-bit large language models — the codebase behind the BitNet b1.58 line of research, where weights collapse to ternary values so models run on CPUs without a GPU. It's back on daily trending as the project keeps shipping: 1-bit embedding models released mid-summer claim up to a 2.28x speedup over full precision on CPU prefill, and a real-time speech-recognition engine built on the same quantization landed in July. This is the practical stack if you want a capable small model on a laptop, a Raspberry Pi-class box, or any machine where a GPU isn't happening.
ppt-master (Python, ~58,000 stars) turns documents or a bare topic into a genuinely native PowerPoint deck — real shapes, transitions, charts backed by your data, and audio narration generated from speaker notes — rather than the flattened slide images most AI presentation tools emit. Sponsored by Kimi and trending again this week at about 58,000 stars, it's the rare AI office tool people actually keep using because the output stays editable in PowerPoint, including your own templates. Reach for it when the deliverable has to survive a human editing it after the AI is done.
andrej-karpathy-skills (about 217,000 stars) is a single behavior file for Claude Code distilled from Andrej Karpathy's public complaints about how coding agents misbehave: making assumptions without checking, bloating abstractions, and touching code nobody asked them to change. Four principles — think before coding, simplicity first, surgical changes, goal-driven execution — in one file you drop into a project, and it remains one of the most-starred repos on the platform. It's the cheapest possible upgrade to an agent's manners: no tooling, no install, just rules the model reads before it touches your codebase.
Worth watching this week.
Free tiers, tiny models, or a behavior file — which of these would actually change how you run your agents? Tell us in the comments.
