Live · 15 of 50 spots left · resets in 6 hours

AI

This week in AI

35 things people built this week, in the order they arrived. Same list for everyone, and it starts again on Monday. How it works

AI · Week 37, 202635/50 spots takenResets in 6 hours
  1. #1⭐ PickMoledeep research agent for your terminalGo CLI research agent that searches sources and backs every claim with a quoted citation, within a set budget.by Lajos Deme
  2. #4Geigersee every AI agent on your machineRead-only CLI that lists the AI agents, MCP servers and extensions on your machine and what each can access.by Atomburst
  3. #6mcp-stamasingle-binary Rust MCP serverDependency-free Rust MCP server with fast grep, git snapshot and Docker watcher tools, built for speed.by Stamatis Manifavas
  4. #7⭐ PickMultiva.Aioffline dubbing for Indian languagesOffline video dubbing for Indian languages that keeps the speaker's own voice and re-syncs their lips.by Mudit Agrawal
  5. #10Vaanivoice translation in 23 Indian languagesNative macOS menubar app for real-time voice translation across 23 Indian languages, typing anywhere.by Jay Patel
  6. #12bhasha-detectIndian language detection in the browserIn-browser language detection for India's 22 scheduled languages plus Hinglish, with no API key.by Kumar Bipul
  7. #14Huzzahcode with AI by editing pseudocodeExperimental editor: write pseudocode, and on save AI syncs it to real source code, keeping your spec as a record of intent.by Daniel Vaughn
  8. #16zoetropewatch coding agents as a live graphRust terminal app that draws a Claude Code or Codex session as a graph of agents, subagents and tools, live or as a replay.by Furkan Kalaycioglu
  9. #18Soupfine-tune LLMs from one YAML filePython CLI and web UI for fine-tuning and post-training LLMs (SFT, DPO, GRPO and more) from a single config file.by Alpamys Makazhan
  10. #20h3.clocal AI video with sound on a MacNative C and Metal engine that runs the MiniMax-H3 model to generate video with audio from text prompts on Apple Silicon.by Salvatore Sanfilippo
  11. #22Ambient Contexta written memory of your workdaymacOS menu bar app that records the text of your focused window as daily markdown files your AI assistant can read.by Cameron Smith
  12. #24LLM Attention Visualizersee what each token draws onBrowser app that runs Qwen3-0.6B locally; hover a generated token to see which earlier tokens it drew on.by Isham Faizal
  13. #26CarWatchan offline AI agent for your carA Raspberry Pi 5 in the car runs a 35B Qwen model, reads OBD data and the owner's manual, and answers by voice offline.by Petrus Pennanen
  14. #28MTPLXfaster local LLMs on a Mac, same outputMac app and CLI that uses the built-in multi-token prediction heads of models like Qwen to speed up local decoding.by Youssof Altoukhi
  15. #30LatticeDBa single-file graph database for RAGEmbedded database that combines graph traversal, vector search and BM25 full-text search over the same data in one query.by Jeff Hajewski
  16. #32modelmapanimated architecture maps of any HF modelPaste a Hugging Face model id to explore its layers, watch a token flow through them, and check if it fits your GPU.by Lizhao Liu
  17. #34Tiny Engineera desk robot that acts out your AI agentOpen-source, 3D-printable ESP32 robot that mimes what a coding agent is doing, like reading, thinking or typing.by Krzysztof Jamroz
  18. #38Your banner hereFree · takes a minute
  19. 14 more open spots#39 to #52
  1. #2mcptoontoken-efficient MCP client128 KB CLI that keeps MCP tool schemas out of an agent's context, so finding tools costs far fewer tokens.by Alex Chen
  2. #3MCP Memorylong-term memory for agentsPython MCP server that gives AI agents long-term memory, stored in Open Knowledge Format and searched with SQLite.by Erfan Reed
  3. #5Graphify C#compiler-accurate code graph for agentsIndexes C# codebases with Roslyn into graphs of calls, references and inheritance that coding agents can query.by Zach Saw
  4. #8Alaapdesign a voice from a descriptionDescribe a character in words and get a reusable voice for it, in English and Indian languages.by Pranav Mishra
  5. #9Swarahear yourself speak Indian languagesSpeak English and hear yourself in Kannada, Tamil, Malayalam and more, in your own voice, on your laptop.by Dhanush
  6. #11RAGgoavoice answers in 11 Indian languagesVoice question-answering in eleven Indian languages, with a 200 ms budget on its fast extractive tier.by Manas Dedge
  7. #13Indiligolearn Indian languages by earVoice-first lessons in 6 Indian languages, with pronunciation scoring that shows which words to fix.by Sagar Kotian
  8. #15backpasstrain your AGENTS.md from real sessionsReads coding-agent transcripts from disk and proposes evidence-backed edits to AGENTS.md and skills, which you approve.by Kun Chen
  9. #17⭐ PickSwiftlet80B Qwen on a Mac in 4.3 GB of RAMSwift and Metal runtime that streams Mixture-of-Experts weights from storage to run large Qwen models on Macs and iPhones.by Leo Nickson
  10. #19RollTabAI piano autocomplete on your iPhonePlay a phrase on a MIDI keyboard and a 125M-parameter model continues your performance in real time, fully on-device.by Simon Edwardsson
  11. #21ThoughtDAGLLM chats as an editable thought graphInfinite canvas where LLM conversations become an editable graph, and the wires you draw decide what the model sees next.by Xia Chen
  12. #23load-bearingmapping how GitHub PRs are writtenClusters GitHub pull request descriptions into ten writing styles by their words, refreshed daily, with an in-page detector.by Louis Abraham
  13. #25Kevin on Kriaa tiny LLM living inside an FPGAA 3.16M-parameter language model kept entirely in the on-chip memory of a $250 Kria KV260 FPGA, with a live chat demo.by Mike Ayles
  14. #27haxa tiny coding agent written in CTerminal-native coding agent in a single C binary, with local models via llama.cpp or Ollama supported out of the box.by Oleksandr Chekhovskyi
  15. #29shoehornquantize a model to fit your exact memoryRust tool that solves a per-tensor mixed-precision quantization so a model fills the memory you actually have.by Bobby Grayson
  16. #31LRU vs agentic KV cachesa reproducible studyPrefix-cache simulator that replays real coding-agent traces to test whether smarter KV-cache eviction beats plain LRU.by Gaurav Agarwal
  17. #33sanoTTS-jpJapanese text-to-speech on a microcontrollerA 559K-parameter Japanese voice model that reads kanji text aloud on an ESP32-S3 microcontroller, with a browser demo.by yousan
  18. #35earthToPixelsreal cities as SimCity-style pixel artTurns aerial map tiles into an explorable isometric pixel-art map, using an image model the team fine-tuned themselves.by Tiago Trindade

15 of 50 spots still open. Put your work on the wall ›