The Stack
10 free AI repos from GitHub created Sept, Nov with under 5K stars: Claude Code agents, local OCR, iPhone control, reverse-engineering defence, each states its own catch.
Ten repositories launched between September twenty-seventh and November third, each under five thousand stars when documented and each openly stating its limitations in its own README. Project Maya runs four mixture-of-experts models on a gaming PC by swapping experts between GPU, RAM, and NVMe SSD, the largest has seven hundred and forty-eight billion parameters, hitting thirty-plus tokens per second on a single RTX fifty-ninety. sand3d lets Claude build 3D CAB models in Python on top of Blender, requiring two to six times fewer tokens than hand-coded Blender Python or CadQuery. Indra is a one-million-parameter OCR model that runs on CPU alone, scoring sixty-six point seven on its benchmark versus PaddleOCR’s fifty-eight point four, reading a page in about a fifth of a second. Kensaku searches one hundred thousand local files by name in eighteen milliseconds with no cloud upload. BLA defends against reverse engineering by bloating builds past REA’s sixteen-megabyte limit, a six-hundred-and-seven-byte Electron archive grew to sixteen point nine megabytes. Kanzo is a kanban board and MCP server in four point seven megabytes that lets you and coding agents share tasks. InferSSD runs a seven-hundred-and-sixty-three-billion-parameter model on a MacBook at just under three tokens per second, about three times faster than plain llama.cpp. earshot transcribes sound for Claude Code agents to reason about, scoring eight out of fifteen in a blind test. API Anything learns request patterns from websites like Google Flights with two example inputs, answering one and a half to two times faster than browser-driven agents. The ESP32 iPhone Agent uses a seven-dollar board as both eyes and hands, streaming the phone’s screen and driving it as a USB mouse and keyboard, it cannot see passcode entry or Apple Pay screens. For builders using Claude Code or local agents, pick by your problem: file search, task sharing, document OCR, API learning, or hardware control.
Chapters:
0:00 Intro
0:31 Project Maya
1:44 sand3d
3:16 Indra
4:46 Kensaku
5:55 BLA
7:13 Kanzo
8:18 InferSSD
9:34 earshot
10:47 API Anything
12:05 ESP32 iPhone Agent
13:37 Verdict
Tools & resources mentioned:
– Project Maya
– sand3d
– Indra: https://github.com/datalab-to/indra
– Kensaku
– BLA: https://github.com/qybaihe/bla
– Kanzo: https://github.com/emanuele-em/kanzo
– InferSSD: https://github.com/wici-ai/inferssd
– earshot
– API Anything: https://github.com/goodnight000/api-anything
– ESP32 iPhone Agent: https://github.com/onehumanbeing/esp32-iphone-agent
About The Stack
The Stack helps you build with AI. Each video takes one tool, model, or workflow and shows how it works in a few focused minutes, with the real benchmarks and real costs.
We go deep on Claude Code and Cursor for AI coding, AI agents and MCP servers, the open-source AI tools and GitHub repos most people miss, RAG and vector search, fine-tuning, and running local LLMs on your own machine with Ollama and LM Studio. We compare models like ChatGPT and Claude, test AI automation with Zapier, Make, and n8n, and flag the tools that actually ship.
Subscribe for new breakdowns: https://www.youtube.com/@the-stack-ai?sub_confirmation=1
#free ai tools #claude code #github
Source
why only links to half of them?
iphone control is the only one on this list i'd actually try. these roundups usually turn out to be the same wrapper with a fresh readme.
00:56 Please don't mock my cat ;(