freshcommits / the list
Every repo, kept.
One trending open source project, three times a day, explained in half a minute. This is all of them.
105 projects15 languagesupdated 3× daily
-
It is not an upscaler. It re-renders the frame your engine already drew, pulling detail out of noise.
-
EverMind-AI/Raven #104
One host drives research, code, design and on-call agents.
-
ninjahawk/livenerf #103
Seventy eight locked questions, rerun every single day, prompts frozen, version pinned.
-
averygan/reclip #102
Paste a batch of links, pick the quality, get MP4 or MP3. No ads, no limits.
-
t8y2/dbx #101
No Java, no bundled Chromium. It makes DBeaver look like bloatware.
-
lnkiai/m3e-canvas #100
Eight thousand stars in a month. No Figma seat, no signup, everything saved locally.
-
dashersw/coyopedal #099
Real amp captures running live, plus gate, drive, delay, reverb and a tuner.
-
mvschwarz/openrig #098
The team is one YAML file, and the agents message each other instead of you.
-
zai-org/ZCode #097
Desktop app, browser, and terminal, one agent writing and running your code.
-
It reads the accessibility tree, not screenshots, so it stays cheap.
-
eval-exec/neomacs #095
Your old config and packages still load, ninety five percent compatible.
-
reladraw/reladraw #094
You write right of and below, so the layout is yours, not Mermaid's guess.
-
It is a hard fork of NewPipe with background music, full playlist downloads and dislikes put back.
-
Hand it an APK or a binary and it picks the tools and runs the playbook.
-
It decides what to keep in three hundred milliseconds, twenty times faster than a model call.
-
Each agent gets an org chart, a task queue, and a monthly budget with hard stops.
-
git-bug/git-bug #089
Issues push and pull like commits. Works offline, needs no server.
-
Runs on your own machine and posts nothing to GitHub.
-
It is a local desktop canvas where Claude Code sketches diagrams you click through into the real code.
-
driceroland/Search #086
Ad blocking and Chrome extensions built in, with no accounts and no telemetry.
-
google/ax #085
One YAML file sandboxes the agent, fences its network, and lets you SSH in.
-
Type slash D C F and it drafts the model, pulling live data from FactSet and PitchBook.
-
xCuri0/ReBarUEFI #083
Patch the BIOS, the CPU finally sees all your GPU memory, twelve percent more frames.
-
dtonon/squalk #082
It runs on Nostr, so your community lives on your relay, not Discord's.
-
volotat/mini-AGI #081
No tokenizer at all. It reads raw bytes and never forgets what it already learned.
-
naw103/foremerge #080
Agents post their intent before writing, so it flags the clash Git never sees.
-
jaredpalmer/kev #079
Feed it a support ticket, get department, urgency and frustration back in half a second.
-
trycua/cua #078
Disposable cloud desktops, so the agent breaks its machine, not yours.
-
mindbox77/zxdesk #077
Overlapping windows, pull-down menus, a file manager and a notepad, all in forty-eight kilobytes.
-
Self hosted, so tax records never touch a paid cloud.
-
Live prices, watchlists and TradingView charts, self-hosted, free.
-
It picks one action and one page element per step. One model call, no chain.
-
It hunts bugs, then verifies each one and deletes its own false positives.
-
asciimoo/hister #072
Find that article you half remember, without Google or a Rewind subscription.
-
OtoDock/oto-dock #071
Each agent keeps its own memory and files, and fires on a schedule or a webhook.
-
ashemag/human-atlas #070
Click any organ or bone, isolate it, or explode the body into every separate piece.
-
It streams the experts off external SSDs, so fifteen thousand in gear replaces two million in servers.
-
timgordontg/engrim #068
Clear the session, and your architecture decisions reload in a thousand tokens.
-
h4ckf0r0day/obscura #067
Thirty megabytes instead of two hundred, and pages load six times faster.
-
Your coding agent writes the HTML, it renders the video. No After Effects, no timeline.
-
aipoch/open-science #065
Twenty four built in connectors let it query PubMed and ChEMBL without you touching a browser.
-
Stremio/stremio-web #064
One interface pulls catalogs, streams and subtitles from addons you install, replacing a shelf of media apps.
-
Hit a hotkey, talk, words land at your cursor. Whisper runs locally, so audio never leaves your machine.
-
humanlayer/skills #062
Install one with npx skills add, then call it as a slash command.
-
ruvnet/ruflo #061
It is the harness, not the model: sandboxes, loops, and memory that recalls what worked last run.
-
anomalyco/opencode #060
It reads, writes and runs code from the command line, and swaps between Claude, GPT or a local model.
-
Magnitude picks the model, downloads it, and unloads it when memory runs short. Ollama makes you guess.
-
It's a skill plus a local proxy: the agent answers tersely, and context gets shrunk before every API call.
-
blader/humanizer #057
It hunts thirty five patterns, delve, tapestry, hollow hedging, and strips them without changing a fact.
-
affaan-m/ECC #056
ECC forces the agent through plan, test, implement, review, verify, instead of guessing.
-
anthropics/skills #055
A skill is one markdown file telling Claude how to do a task, loaded only when it is needed.
-
Hermes from Nous Research runs on a five dollar server and imports your old OpenClaw keys and memories.
-
VoiceStudio packs sixteen speech engines and eleven transcribers, six hundred languages, all running offline.
-
It ranks first on fev bench across a hundred tasks, using just two hundred million parameters.
-
fmtlib/fmt #051
It's called fmt. Python style format strings, checked at compile time, so a bad placeholder fails the build.
-
Deep research, drafting, and a seven agent peer review, wired into one pipeline with citation checks.
-
iv-org/invidious #049
It scrapes YouTube directly, so your subscriptions live on your server, not your Google account.
-
Gitlawb/openclaude #048
Same terminal loop as Claude Code: bash, file edits, grep, MCP tools, streaming.
-
Osmantic/ODS #047
ODS installs Ollama, Open WebUI, a vector database and n8n workflows as one stack, detects your GPU, and picks models that fit it.
-
jingyaogong/minimind #046
MiniMind trains a language model from scratch in plain PyTorch for about three dollars.
-
majd/ipatool #045
It signs into your Apple ID, grabs the license, and pulls the package. No Mac, no iTunes.
-
Slash last thirty days researches any topic across a dozen sources from the past month, scoring by human engagement instead of S E O.
-
unclecode/crawl4ai #043
Crawl4AI is an open source Python crawler that turns any website into clean, LLM-ready Markdown for RAG pipelines.
-
Lakr233/vphone-cli #042
Vphone wraps Apple's own Virtualization framework and patched firmware, so one command boots a working virtual iPhone on Apple Silicon.
-
bigskysoftware/htmx #041
htmx turns HTML attributes into AJAX so the server's HTML swaps straight into the page.
-
THU-MAIC/OpenMAIC #040
Give it a topic or upload your own materials and it builds slides, quizzes, and interactive simulations in minutes.
-
tailscale/tailcat #039
Netcat over encrypted WireGuard with no Tailscale account or root needed.
-
Screenshot to code turns any screenshot, Figma frame, or screen recording into working React and Tailwind you run locally.
-
It scans your repo and maps every call chain and dependency into a graph, then serves it to Claude Code, Cursor, or Codex through seventeen MCP tools.
-
Scientific Agent Skills is a pack of one hundred sixty three skills for any agent that uses the open Agent Skills standard.
-
Persistent memory for your coding agent, installed with one command.
-
It pulls public feeds, live flight telemetry, ship positions, earthquakes, NASA fire data, onto Google's three D map tiles.
-
OpenMontage runs twelve production pipelines: script, voiceover, stock footage, and final render.
-
An AI agent that drives a real browser, replacing brittle Selenium scripts.
-
It forces a laziness ladder: skip the feature, reuse code that already exists, use the standard library, then write new code last.
-
tt-a1i/archify #030
Archify is an agent skill for Claude Code, Cursor, and Codex. Install it once, then ask it to map your repo.
-
Instead of a vector database you cannot read, every fact it learns becomes a local file you can edit or delete.
-
Claude obsidian ingests papers and web pages into your Obsidian vault as linked notes where every claim carries a source citation.
-
TradingAgents runs a multi-agent LLM trading firm you can pip install.
-
makeplane/plane #026
Plane is a self-hosted Jira replacement with sprint cycles, docs and analytics built in.
-
A Claude Code framework that scrapes jobs, tailors your CV in LaTeX, and preps interviews.
-
Use Claude Code, Codex, Pi, and OpenCode for free (1.3B+ free tokens) from your terminal, app, IDE, or phone like OpenClaw (voice supported + ToS friendly)
-
apache/maka #023
Apache Maka (Incubating) is a local-first AI agent workspace. Model messages, tool calls, tool results, permission decisions, and termination events are recorded as an append-only log.
-
block/buzz #022
A hive mind communication platform
-
AprilNEA/OpenLogi #021
⚡️A native, local-first alternative to Logitech Options+, written in Rust 🦀 — remap buttons, DPI, and SmartShift over HID++. No account, no telemetry.
-
openai/codex #020
Lightweight coding agent that runs in your terminal
-
modular/modular #019
The Modular Platform (includes MAX & Mojo)
-
jundot/omlx #018
LLM inference server with continuous batching & SSD caching for Apple Silicon — managed from the macOS menu bar
-
obra/superpowers #017
An agentic skills framework & software development methodology that works.
-
mattpocock/skills #016
Skills for Real Engineers. Straight from my .agents directory.
-
Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
-
Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors
-
local multi-agent harness
-
Production-grade Rust-native trading engine with deterministic event-driven architecture
-
usestrix/strix #011
Open-source AI penetration testing tool to find and fix your app’s vulnerabilities.
-
利用 AI 大模型和自动化工作流，根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.
-
ToolJet/ToolJet #009
ToolJet is the open-source foundation of ToolJet AI - the enterprise app generation platform for building internal tools, dashboard, business applications, workflows and AI agents 🚀
-
OpenCut-app/OpenCut #008
The open-source CapCut alternative
-
Two-way communication bridge between Claude Code sessions using the Channels API
-
basecamp/omarchy #006
Beautiful, Modern & Opinionated Linux
-
MakazhanAlpamys/Soup #005
Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU.
-
unslothai/unsloth #004
Local UI to run and train LLMs and diffusion models, including Qwen3.8, Kimi K3, MiniMax-H3, Gemma 4, DeepSeek-V4, FLUX and more.
-
megadose/holehe #003
holehe allows you to check if the mail is used on different sites like twitter, instagram and will retrieve information on sites with the forgotten password function.
-
14MB foundation model for tiny devices; phones, wearables, smart home, and robots.
-
38 editorial diagram types for Claude Code, Codex, and Pi. Self-contained HTML + SVG. No shadows. No Mermaid slop.