ByteAsk
Open source AI coding agent for C and C++ in the terminal. Free plan.
Best AI Builds on X
Models, wrappers and small products that do something a frontier model could not do last year. We only list ones you can try, not launch-thread vapor. 286 builds on this page, each with the original post.
Thimble
Open source Claude Code plugin that explains what an agent did, with visualizations in the terminal.
Calibrain
Open source project that runs a classification model against crypto perpetual markets on a testnet.
SuperMimic
Open source code for SuperMimic, a humanoid motion controller from Hybrid Robotics at UC Berkeley.
Claude Fable 5.2
Anthropic's public model for the hardest technical work. Up to 30% faster than Fable 5.1, and cheaper. Live on the Claude apps, Claude Code, and the major clouds.
Claude Mythos 5.2
The restricted twin of Fable 5.2. Access is through Anthropic's cyber and life sciences verification programs.
Fable 5.2 on AWS
Fable 5.2 on Amazon Bedrock and Claude Platform on AWS. First-party from AWS.
Which Claude model to build with
Anthropic's guide to picking Fable 5.2, Opus 5.5, Sonnet, or Haiku, and when to raise effort instead of switching models.
Phone Use
An agent watches a real iPhone over AirPlay and drives it over Bluetooth. No app on the phone. Open source.
Timon
Open source agent workstation. Memory, routines, and a browser. A terminal, kanban, git, and Docker when you need them. AGPL.
Dim
Open source agent runtime. Plugins, and an event timeline the agent works along.
Voidless
MCP that builds a Godot character from one chat: directions, a walk, a spell, and the effect.
Open Cinema Studio
Local film studio on ComfyUI. Write, direct, generate picture and sound, then edit. Open source.
Ratchet Loop
Claude Code and Codex loop that runs the tests, then has a second model try to break the result. Open source.
Fastlane
Learns a brand, makes ads from what is already winning, and launches them.
Checkride
Scans MCP servers and agent tools for actions that run without asking. Open source.
Host-Caster
AI radio that talks about your calendar between songs. Self-hosted, with a site to try it.
Agent Desktop
Computer use for a coding agent, on macOS, Windows, and Linux. Each agent gets its own cursor. Open source.
ForgeKernel
Open source harness for writing and comparing GPU kernels. The published set includes a DeepSeek FlashMLA prefill kernel.
Local Code
Private coding assistant for Mac, iPad, and iPhone. It runs a model on the device. Open source.
Keyfit
Checks that Claude and Codex are on the right account before a project starts. Open source.
Autopilot
Agent that plays a game and writes up the bugs it hits. Unity, Godot, and Unreal. Open source.
MotionBridge
Turns an AI video into a skeletal animation on your own rig. ComfyUI nodes. Open source.
Mogfield
Open source AI film studio on Comfy. Storyboards, restyle, and character boards, local or in the cloud.
Revideo
Browser editor for short videos, with scripting and discovery. Open source.
RSS Agent
Literature search agent you can point at the papers you care about. Open source.
Sushi
Runs quantized models on Apple Silicon, in STQ and EXL3. Open source.
Postiz-cheap
Self-hosted social scheduler with an AI post writer and a content calendar. Open source.
REA
One MCP for reverse engineering binaries, apps, and runtime behavior. This release adds environments for training local models. Open source.
DOOM-SED
Interactive demo of an open-vocabulary sound event detector. It scores labels together so similar sounds collide less.
Temporal Fusion Price Model
Multi-timeframe price model built with a temporal fusion transformer. Open source.
Open Math Harness
Runtime for agents doing math research. Open models, scientific computing tools, and Lean autoformalization. Open source.
Xemo
Describe an onchain product in one prompt and it ships as a live business, written and deployed in a sandbox, then charged through Stripe and x402.
Cyberdyne T800
An open source MIT agent with a body, vision, hearing, and pain. The author uses it as stress relief when a model breaks his flow.
Miks
A terminal for running a bunch of coding agents at once, with workspaces in a sidebar. Made as a replacement for Terminal.app.
Stations
Ankor reads your X profile, fetches stories, writes a script, and plays the edition back with source links.
LeanREA
An 8 KB skill for reverse engineering with plain CLI tools and no MCP server. Free and open source.
Object Anatomy
A gallery that explodes everyday objects so you can read what each part does.
Keept
Pay-on-delivery escrow for AI agents. The agent holds the payment and releases it when the work arrives. A PayBox plugin.
Sparky Command Center
A command center for a home AI lab: temps, watts, memory, and live throughput across Spark boxes and a 3090, plus a chat that reads the board. Open source Python.
Stall
A shop you open by chat, so WhatsApp sellers can be found and paid by AI assistants. Built for the PayBox hackathon.
gann
A driving test for AI agents with wallets. Agents get a small budget, deceptive merchants, and prompt injections, then an auditable report. Built for the PayBox hackathon on Base.
Frametell
An open source director agent. A sentence or some notes go in, and a narrated explainer video comes out.
Vellum
An open source diagram editor for scientific drawings. Draw, annotate, and the edit is done.
DeepSeek price clock
A Chrome extension. The toolbar icon is green in DeepSeek's off-peak hours and red when the price is peak.
Seekter
A Claude Code agent that finds job posts and fills the applications in your own browser.
Eddie
A native Mac video editor with an AI color grade for DJI D-Log M footage, and an MCP server so an agent can edit too.
RSI-Inference Bench
Benchmark for agents working on real vLLM and SGLang tasks: day-0 model support, inference optimization, and bug fixes. From Apple ML Research.
Muxin
One app for every computer: run coding agents on Macs, Linux, Windows, and cloud machines from a single window. Free and open source.
Circle to Fix
Circle a bug on a page. The prompt you paste into Claude Code, Cursor, Codex, or ChatGPT includes the screenshot and the code behind it.
Scenario for Blender
An open source Blender plugin that puts a 3D team in the editor: concept, model, rig, materials, cameras. Claude Code can run it through MCP.
The Grill
Measures whether a change made a local model server faster, slower, or too close to call.
OpenUI
Generative UI for agents. The model streams an interface from your own components, not just text. Version 1.0.
ERP AI
Describe a business system and it builds the app. Proto, an agent, then runs the records and workflows inside it.
People vs your drawing
Draw something. Jev guesses what it is, then judges the drawing.
Parametric
An open source parametric CAD app from Puter, with an AI assistant that can build the model.
Fieldtwin
Local stand-ins for the Jobber and QuickBooks Online APIs, so you can test integrations and agents against a fake business.
EntroPack
PyTorch tensor compression that quantizes a model to whatever bit rate fits the GPU.
Sound Dictionaries
Pronunciation practice that roasts you and flirts with you, in Chinese, Japanese, Spanish, and Korean. Voices from Fish Audio Drama 3.
Storebound
Open source toolchain that takes a mobile app from an empty folder to Submit for review on the App Store and Google Play. It stops at the submit button.
Billu
Talking-cat toy on Sarvam Bulbul V4 Flash. Chat, repeat after it, or throw a tongue twister. 83 voices, 11 languages.
AgentSec
Open source adversarial tests for AI agents: prompt injection, tool misuse, memory poisoning, and MCP scanning.
Lexis
Plain English in, a shell command out. A small model on your machine, and it asks before anything risky.
Haiku Harbour
A seaside town you share with hundreds of NPCs and other players. Built with Claude Haiku 5.5.
Atlas Camera Studio
Camera controls for generated images and video. The studio behind a WorldLabs clip, now open source.
bend-ml
Machine learning in Bend 2. A bad matrix shape is a type error. GPT-2 small matches PyTorch token for token.
Claude Opus 5.5
Anthropic's first Claude 5.5 model. Fable-level on most work, about 40% cheaper than Opus 5. API id claude-opus-5-5.
Stickman (Opus 5.5)
Stickman fighter with character customization and skills. One HTML file from Opus 5.5. Playable.
netherspace
Low-effort platformer. An AI agent wakes up and reveals the world as he solves queries. Claude Opus 5.5.
laya-apple
Same Laya weights on MLX GPU and Apple Neural Engine at once. P99 1593ms to 80ms on M4 Max.
Halo
Post-training framework for open-source models. Up to 2.8x TRL throughput, less peak memory, weights stay HuggingFace-native.
openJev Verdict
Open 151M decision model on ModernBERT. JevBench #2 on public tasks after the inference-engine fix. Weights and a WebGPU playground.
Muse on agentic commerce
Asked Muse to research itself, Amazon vs Shopify, and which stocks win if agents start shopping. It built the page.
Laya
Open Apache 2.0 System One model. ModernBERT, typed choice/score/noul in one pass. Fine-tune notebook on Kaggle T4s. Not Jev.
Kev
Jared Palmer's open Jev-like family on Qwen3.5. 0.8B, 4B, 9B. TypeSafe SDK with a base_url change. Train on one H100.
SemIf
Open decision model that reads option logits from local Qwen-class weights. Browser demo at openjev.com. TypeSafe SDK compatible.
laya-mlx
Laya on Apple MLX. Under 1GB RAM, 7 to 14ms on M3 Max. Snake at 60 decisions a second, no PyTorch, no cloud.
APUS-OpenJev
4B, 9B, and 35B-A3B decision family for browser agents. ~26ms median. TypeSafe SDK. Weights on Hugging Face.
ballot
Open System One model in the same class as Jev, Kev, and Laya. State plus typed questions in, calibrated probabilities out.
MetalML
scikit-learn workflows on Apple Metal GPU. Change the import, keep fit/predict. Falls back to CPU when it has to.
Muse
Meta's personal AI agent. Own Secure VM, own browser, does the work across apps. Powered by Muse Spark 1.3.
Muse Spark 1.3
Meta Superintelligence Labs flagship. Agentic and coding model in Muse Code and the Meta Model API.
SeaCat
A system one model like Jev. Typed answers with probabilities, no generated text. pip install seacat or npm install seacat-ai.
Triage Bot
Realtime support triage with Jev routing tickets, then Cerebras drafts the reply. Typed decisions, no text to parse.
LocalVisionAI for Unity
On-device LLM image understanding in Unity. Offline OCR by describing the target in natural language.
LocalJev
Poor man's Jev on local models via omlx. Same API shape as Jev, for people without access yet. From GitHub Next.
YouTube Sponsor Detection
Chrome extension that listens to YouTube audio, detects sponsor segments, and skips them in real time. Open source, BYOK.
WEKTOR
Local-first AI data analyst. Your data stays on your machine. Only column stats leave.
Jev
TypeSafe's System One model. Typed decisions, no chat. Trained with RLCD. Output tokens free.
Ask Jev
Playable Jev demo. A needle that panics in realtime. Possibilities, not text.
Weeping Angels
Browser horror game: they only move when you cannot see them. Desktop and headphones. GPT-6 Astra plus fal.
Jev Trader
Trading bot with Jev. Buy or sell from a price feed, then places orders on Monad via Kuru every 300ms.
NEMA
302-neuron C. elegans wiring trades simulated SOL. Silence a neuron and replay the market.
The GARCH Method
Open-source GARCH trading framework: Pine Script indicator, agent skills, and a Claude plugin.
SealGate
Local LLM gate that hides sensitive info before it reaches online model providers.
DecisionBridge
Bring Jev-style decision scoring to existing LLMs. Calibrate on your data, set a review threshold, no fine-tuning.
Sci-Fi Fetch Quest
Gaussian splat browser game in a World Labs Marble scene. three.js, nav mesh, light probes. First-party, with a walkthrough.
iPhone Duo
Two Astra prompts, a reference video, and a product page. Playable React fold demo.
Fly With Me
Sunset to nightfall to moonrise to Milky Way, all JS, no images. First-party Astra build.
DexGPT
One video of a human hand becomes a real2sim pipeline. Tracking, IK to a 44-DoF hand, grasp refine. First-party, OSS.
S3 PX4 Flight Controller
Open-source FPV flight controller. ESP32-S3, EasyEDA project and gerbers. Designed with Codex and GPT-6 Astra.
EgoWAM
World action model that lets robot policy scale with in-the-wild human video. CoRL 2026. Code and data are public.
Sol-H3
NVIDIA inference stack for MiniMax-H3. Five seconds of video in 1.65s on 8x B300. Faster than playback. Apache 2.0.
Open Steps
Free agent skills that force a straight verdict, plain steps, and a next action. Built by a non-engineer running a software company.
ComfyUI Reactor Connector
Open-source ComfyUI nodes and workflows for Reactor endpoints, including MiniMax FastH3.
LiveAvatar GPT-Live Demos
HeyGen open-sourced real-time AI avatar demos with OpenAI GPT-Live-1, LiveAvatar, and HyperFrames.
vLLM MXFP4
MXFP4 vLLM build for Radeon AI PRO R9700. NVFP4 Qwen3.8 27B at 5809 tok/s prefill.
Social Quack
Buy-once social scheduler. Unlimited channels and posts. Talk to it from Claude or ChatGPT.
One Face Lock
GPT Astra / Claude skill that locks one face, voice, and light across many videos.
Live Classroom
Generative video classroom on fal H3 Max. Ask a question, watch an animated explainer, queue follow-ups. OSS, add your own keys.
Rattlers
Turns a favorite song into a custom blind-box toy. Built with Reve. Playable.
x-algorithm
The open-source X ranking code. Daily updates, and the first public PR is now live on X.
Pangram Badge
A live Pangram AI-detection badge you can drop on a site. First-party, with a generator and the repo.
Code-as-World
Agents turn physical experience into executable world code. Models, paper, and a project page from MirroS.
Claude Fable 5.1
Anthropic's public Mythos-class model. Long-horizon coding and research, cheaper cache reads. Live everywhere today.
Claude Mythos 5.1
Same weights as Fable 5.1, fewer safeguards. Restricted to cyberdefenders and life scientists via trusted access.
Fable 5.1 on Google Cloud
Fable 5.1 on Google Cloud Agent Platform. First-party from Google Cloud.
Claude Security on Mythos 5
Claude Security scans now run on Mythos 5. Public beta for Claude Enterprise. First-party.
Mythos cryptographic weaknesses
Mythos Preview found new attacks on HAWK and a reduced AES. First-party research from Anthropic.
OpenMythos
A from-scratch PyTorch reconstruction of what Claude Mythos might look like. Looped transformer, MoE, MIT.
Cyphral Distich
Fable 5.1 decoded Sir Thomas Urquhart's 1653 cipher, unsolved for 373 years. First-party from Vals AI.
spring-ai-agents-md
Zero-config Spring Boot starter that adds AGENTS.md support to Spring AI apps. First-party.
LabelOne
Local-first annotation workspace. Non-destructive image pipelines, synced views, and model feature inspection.
GLM-5.3-Flash
Z.ai's natively multimodal Flash model. 320B total, 18B active, 1M context, MIT weights. Previewed as Ox Alpha.
GLM-5.3 open weights
The full GLM-5.3 coding and cyber-defense model is now downloadable on Hugging Face. Same base as 5.2, post-trained.
Qwen3.8-Flash-Next
Alibaba's multimodal MoE and a preview of Qwen4. 125B plus 51B n-gram embeddings, 6B active, 262K native context. Open-weight.
GLM-5.3-Flash GGUF
Unsloth 3-bit GGUF so GLM-5.3-Flash runs on 128GB RAM. First-party local pack.
GLM-5.3 GGUF
Unsloth 2-bit GGUF of the full GLM-5.3. 1.51TB down to 239GB, for a 256GB Mac or mixed RAM/VRAM.
DeepSeek Harness
Official MIT harness from DeepSeek. Everything is a plugin: models, tools, sandboxes, loops, and UI. Run with npx @deepseek-ai/dsh web.
Amplio
DeepMind's harness for autonomous research runs that last days to weeks. Used internally on Simply, now public.
Dot Reflex
Inference runtime, local API, agent loop, and training provenance for Dot's 14B model. Company open-sourced the engineering layer.
HRFFA
High-angle robust fast face alignment. Whole-head dataset and training pipeline, first-party from PINTO.
Hermes plugin for BackSearch
Teknium's stand-alone Hermes Agent plugin. Search the web as it was on a given date, like a Wayback Machine for agents.
Finite Systems
AI-native CAD in the browser. The author designed a physical coaster in it, then manufactured it.
LLM cliché highlighter
Simon Willison's page that paints LLM tells. 38 patterns now.
Cyclo Control pick-and-place
ROBOTIS OMY arm: Gemini reasoning, GraspNet 6-DoF poses, collision-free pick-and-place on ROS 2.
Summit Scout
All 63 US national parks scored on remoteness, scenery, elevation, biodiversity, plus camping and directions.
AO
Desktop app that runs Claude Code, Codex, Cursor, and 26 other coding-agent harnesses in one place.
chock
Sandboxed AI harness in Zig. Landlock, eBPF, and namespaces isolate tool calls. WASM plugins over IPC.
Surflo
Turn any number of photos into one 3D surface with Flow Matching. Code and model are public.
rome-apps
Internal agent apps Rome runs the company on, open-sourced so you can see how they actually work.
claude-openrouter
Installer that puts GLM-5.3 next to Claude models inside Claude Code, via OpenRouter.
GitNexus
Turn a repo into a knowledge graph and hand it to Codex, Cursor, or Claude Code over MCP. Runs in the browser.
Redact
Blur faces and objects on a Mac without uploading the video. YOLO on the Neural Engine, tracking so a missed frame does not leak a face.
Composer
Generative art where the model writes JavaScript in an isolated worker instead of getting a tool per shape.
live-visuals
Speaks numbers and processes, draws charts and diagrams while you talk. Local Whisper. Cannot invent a number you did not say.
HTD whole-body controller
Decoupled humanoid whole-body controller. Train on one GPU, drive it in the browser, zero-shot onto the robot.
DROID+
NVIDIA SRL's robot control stack, rebuilt so the hardware is actually easy to set up. pi0.5 doing the hello-world task.
OpenTag
Tag an AI in Slack the way Claude Tag works, but pick any model. Live with 10 teams.
PointDiT
Pixel-space Diffusion Transformer for dense 3D point maps. Code, weights, and training pipeline are public.
God's Eye View
Spy-simulator in the browser on real data: planes, ships, satellites, traffic cameras. Talk to the planet. MIT.
hermes-qwen-vllm
vLLM plugin that fixes compaction and thinking-mode bugs when running Qwen 3.8 27B as a local Hermes agent.
kimi-chip
RL harness that treats ASIC design as a hill-climb. RTL and GDS are public. Write-up at luoluo.ai.
Beacon
iOS app to download open models from Hugging Face and chat with them on-device. No API keys, no terminal, no subscription.
omaharness
Thin harness so an agent can drive a full Omarchy desktop: Hyprland, AT-SPI, real browser, files.
Super Vault
Self-hosted knowledge vault for Hermes: ingest articles, YouTube, RSS, bookmarks, then search them from the agent.
Proliferate
Open-source AI IDE for Claude Code, Codex, Grok, Cursor, OpenCode. Parallel agents, each in its own worktree.
Codex Usage Menubar
Tiny macOS menu bar that shows how much Codex you have burned without digging through menus.
cubicle
Harness that gives every AI bot its own directory, browser profile, cookie jar, and environment. Grants expire; the door closes.
sparktop
Dashboard for multi DGX Spark setups: inference throughput, GPU processes, and more.
RulesAsPrograms
Turns each agent rule into a function that checks the agent's work, instead of hoping it follows the prompt.
Spreadsheet Webcam
Webcam emulator inside a spreadsheet, built in Google AI Studio on Gemini 3.7 Flash. Export images and CSVs.
Slopdar
Open-sourced detector for AI-slop websites. Add a pattern, improve a detector, or build on top of it.
Qwen3.8-27B on RTX 3090
Inference stack for Qwen3.8-27B on a single RTX 3090: 195k context, 82 tps single request, 672 tps peak.
Qwen3.8-27B NVFP4 on DGX Spark
Day-one Qwen3.8-27B deploy: NVFP4, 262K context, one DGX Spark, one command. Full reproducible stack.
Qwen3.8-27B FP8 on 2x 3090
27B on two consumer 3090s at ~75-85 tok/s with NVLink. Tool calling, reasoning, vision, MTP. Full deploy open-sourced.
PARS
Poole Adaptive Reasoning Stack as a skill: generate exact binaries, research, and audit with PARS.
TencentDB Agent Memory
Open-sourced agent memory. Mid-session compression cut 61% of token usage in their case.
agent-memory
Open agent memory on Cloudflare Workers and Durable Objects. Profiles, extract, recall. One-click deploy. MIT.
Treg
OpenRouter for tools: 2,600 agent-friendly APIs, pay per call, 0% markup, search by task. Open source.
Cook the Dungeon
First-person roguelike deckbuilder: defeat monsters, harvest parts, cook meals. Playable demo after two months of building.
handMediaPipeHands
85-line finger-tracking toy for Physical AI video cleanup. Claude Code plus MediaPipe.
Grand Theft Austerity
Open-world Bucharest satire from an AI Game Jam. Playable plus repo.
Open Cowork
YC-backed open-source computer-use alternative to Claude Cowork.
HyperFrames
Open-source motion-graphics framework and skills from HeyGen.
mind
Persistent local memory for coding agents. One Python file, no dependencies, explainable recall.
HumanCLAW
Can a VLM act through a body? Open-source benchmark for action intelligence, decoupled from motor control.
Benchkit
25 small biology apps for day-to-day experiment planning, built in a couple of hours on GLM-5.3.
Local Knowledge Graph
Reason around a question without context-window limits, and visualize the knowledge graph.
Minovative Mind
AI coding tool with Gemini 3.7 as the default. Bring your own Google AI Studio key.
expert-stream
Public engine to stream DeepSeek V4 Flash experts on an 11GB laptop. LRU under one layer is 0% hits, so they published it.
shadcn Chatbot Template
Minimal chatbot template: new components, typeset, tool calling, HITL questionnaire, message parts. One-click deploy to Vercel AI Gateway.
omp-best-of
oh-my-pi extension for best-of-N + self-verify: sample several solutions, rank them with the same model as verifier.
Muon
Shared local brain for Claude Code, Cursor, Codex and OpenCode. MCP plus CLI so agents stop redoing each other's work.
Flutter-QNN
On-device LLMs in Flutter on Snapdragon. Gemma and Qwen via Qualcomm QNN, Hexagon NPU, local streaming.
dbrain
Personal search over X, YouTube, RSS, Bluesky and Mastodon. Web UI plus MCP so agents can query it.
SkinAI
iOS app that scans your face, scores your skin and gives a shareable routine.
Local Qwen Breakout
A single-file HTML canvas Breakout game written end to end by a local Qwen 3.8 27B through Hermes.
Shoply AI Search
AI search, product recommendations and customer support for online stores. Ranked second for AI Search on the Shopify App Store.
StockScans Research AI
Research assistant for investors, built to dig through company filings and documents for deeper insight.
termpeek
MIT tmux pane that renders what your coding agent can see and you can't: images, video, PDFs, diffs and X cards. Bash 3.2, nothing leaves the machine.
Mac Developer Bridge
MCP bridge so ChatGPT can operate your real Mac: shell, files, PTYs, background jobs and read-only Codex history. MIT. No model calls.
BRIKS
AI app for SMBs to chat with customer feedback, conversations and sales. Launched on Product Hunt.
Scrimba Explain
Turns any question or block of text into a video explainer instantly. Also ships as a Chrome extension and as Claude and ChatGPT plugins.
b3t Protocol Launchpad
A custom token launchpad paired with a Claude pipeline that turns study notes into textbook-style blog posts.
Claude Code Trading Bot
A trading bot built with Claude Code in two days that scans 50+ markets for arbitrage. The post claims it turned $68 into $750k.
Local AI Box
A $2,300 local AI rig: two RTX 3090s for 48GB of VRAM, a dual-Xeon board and 28 cores, running 70B quantized models fully offline.
Vibely
Closes the gap between the idea in your head and something people can actually use and ship.
AI Market Dashboard
Continuously tracks the biggest AI stories, funding rounds, partnerships, model releases and infrastructure news.
REDCELL
Open-source AI agents that run a full pentest end to end, from recon through exploit, reverse shell and pivot, then write a client-ready PDF report.
AI Cowork Harness
An AI cowork harness tested heavily against open-weight models. Runs well on DeepSeek V4 Flash and Meta’s Muse Glimmer 30B, which is agentic out of the box.
AI Tumor Board + Patient Agents
Two cancer tools from a practicing oncologist: an AI tumor board for clinicians, and a multi-agent patient-facing tool that answers the questions doctors have no time for.
Inline Prompt Optimizer
Fires a forked thread and returns an improved prompt, with a diff view component to show what changed.
Interactive 3D Human Heart
Human heart anatomy site built with Gemini 3.7 Flash. Explore the heart in 3D and learn what each part does.
Merge Agent Handler
An agent built with Claude, GitHub and Notion that finds recently merged PRs, summarizes what shipped, and posts a fresh digest to Notion every Monday.
Magnific Auto Layers 2.0
Splits any AI-generated image into editable layers for subject, text, logo and background, so each piece can be restyled independently.
Install Instructions Agent
Generates software installation instructions for application owners and packaging teams, cutting the manual documentation work.
Snaptrude Apps
Turns a firm's design workflows into software. Describe the workflow, define the rules, and get an app that runs on your building model.
Japanese Tofu Drift
A 3D Japanese tofu delivery drift game built entirely on Crayon Pro, an AI game engine.
GLM-5.3
Z.ai’s new coding and cyber-defense model, post-trained on a 743B base. Agentic coding jumps over GLM-5.2 and it is live now via GLM Coding Plan and ZCode.
OpenRec
Free, open-source Granola alternative that actually looks good. Bring your own OpenAI and AssemblyAI keys, optional R2 storage.
LedgerPDF
Free, offline, open-source workpaper binder for accountants. Agents connect over local MCP, every change is journaled, and humans still sign off.
Mycelium + Forage
Agent tools for any SDK that live-refresh and can be searched or described, plus a durable agent mailer along the lines of Oban for JS.
InventList Weekly Launches
Ten fresh indie launches in one roundup: Procright, Anomaly Ai, Bloom, Popform, Max, Bon Split, Archmaster, Flexify Suite, Kage and Roman Names.
pgbot
Free open-source Postgres intelligence for agents and apps. Finds unused indexes, slow queries and drift. Written in Go. No dashboards.
Agent with Persistent Memory
Fully shipped AI agent with persistent memory built in less than a day.
Meetily – Local Meeting Assistant
100% open-source meeting assistant built with Rust + Tauri. Transcribes 4x faster with Whisper/Parakeet and runs completely locally.
Book-to-Skill
Turns any PDF, EPUB, DOCX or Markdown book into a ready-to-use Claude Skill that loads only the relevant chapters when needed.
ds4 – DeepSeek V4 Inference
Native inference engine optimized for DeepSeek V4 and GLM models with support for Metal, CUDA and ROCm.
ThoughtDAG
Local-first visual LLM workspace where graph edges control model context, with branching, merging and document extraction.
PentesterFlow
Open-source CLI agent for authorized pentests. Plan, act, observe, verify, report. Local or hosted LLMs. Burp + MCP.
Grok + Brain Sensing App
Grokathon project that lets people speak without using their mouth by combining brain-sensing tech with Grok Voice.
Grok Build Power Features
Deep dive into the lesser-known but extremely powerful features SpaceXAI has shipped in Grok Build.
Inverted Lambda Teleoperation
Human-in-the-loop teleoperation infrastructure aimed at creating an open, distributed learning network for robots.
YOTS – Grok-based Agent
Persistent Grok-based AI agent with its own wallet, identity and ongoing lore around the “Year of the Singularity”.
Fabric-Aware Virtual Try-On
Upload your photo and measurements to see exactly how clothes will look and fit on your body.
18-Persona Deliberation Council
Open-source repo that grew from an 11-agent experiment into 18 reasoning personas with multi-provider routing and confidence-weighted verdicts.
Claude Code Job Interviewer
Program built with Claude Code that interviews job applicants over chat, scores fluency and thinking, and flags likely AI-assisted answers. Live in production.
Eliza / elizaOS
Fully open-source agent operating system that runs on mobile, desktop and web with voice + chat control.
AI Interior Redesign Tool
Vibe-coded tool that redesigns your room using real furniture you can actually buy, with a full shopping list.
AI Assistant with Phone Number
AI that has its own phone number and email. Makes calls, waits on hold, and books follow-ups. #1 on Product Hunt.
Talk-to-Code Agent
Coding agent you speak to instead of typing. Say what you want and it ships in the background.
Text-Only Scheduling Agent
Agent that schedules meetings and books reservations 100% over text message. No app required.
Gemma Offline Translator
Fully offline portable translator running Gemma on a Raspberry Pi 5 inside a custom 3D-printed case.
Minutz – AI Meeting Intelligence
AI tool focused on meeting intelligence and action extraction.
Saral AI – Voice Agent Platform
Voice agent platform for building conversational AI experiences.
Kimi K3 in Pure C
176KB C99 engine that runs the 2.78 trillion parameter Kimi K3 on a single CPU with 8GB RAM by streaming experts from NVMe. No GPU, no framework.
Prime Agent
Open-source self-improving coding harness where sub-agents are function calls in a persistent kernel. Scored 95.5% on ARC-AGI-3, above the human expert baseline.
Cloudflare OS
Kenton Varda's remake of his startup Sandstorm as an AI agent workspace with connectors, built entirely on Cloudflare Workers. #1 on Product Hunt.
Voice AI Sales Coaching Product
Voice AI product that gives live coaching reports (talk ratio, objection handling, buyer verdict). Reached Top 5 Product of the Day.
anydoc by Firecrawl
Rust open-source parser that converts PDF, DOCX, PPTX and 10+ formats to clean Markdown in under 5ms, built for agents. 12.5k stars in days.
Superblocks 3.0
Secure enterprise-grade platform that lets companies vibe-code production software inside their own AWS environment.
Forge – AI Project Shipper
AI coding agent that takes a single prompt and ships a complete project.
Xiaozhi – ESP32 Voice Assistant
Fully open-source framework that turns an $8 ESP32 into a real-time autonomous voice assistant with local wake-word detection.
Prismor
Open-source control plane for AI agents. Defense by default. Used by thousands of devs a month.
Migma.ai
AI product frequently highlighted in maker shoutouts for its capabilities.
QM by Y Combinator
Tool highlighted by makers for its usefulness in the current AI building wave.
28.9M Parameter AI on ESP32
Tiny open-source AI model running fully offline on a few-dollar ESP32-S3 chip at ~9.5 tokens/s. No cloud needed.
Polygres – Billion Token Context
Open-source tool that gives frontier models effective 100M–1B token context by using a graph-based retrieval system.
Axis Robotics Data Engine
Full Physical AI data collection and processing platform for training robots from simulation and egocentric video.
ESP-32 Zero-Latency Auto Aim
Custom local AI running on an $8 ESP-32 that tracks humans with zero computational delay.
Text-to-CAD
Open-source tool that turns plain English descriptions into production-ready CAD files (STEP, STL, URDF, etc.) for robotics and mechanical design.
OpenWorker by Andrew Ng
Open-source desktop AI coworker that delivers finished work: polished documents, Slack messages and calendar updates, with 35 connectors and permission tiers.
Open Source AI 2.5D Game Engine
Game engine built specifically for AI agents. Small enough for agents to understand the whole codebase and generate complete playable games.
![Screenshot of a writing-analysis tool showing paragraphs of text with phrases highlighted in two shades of yellow, brown circular "3" badges, and a black tooltip overlaying the first line reading: "No X, no Y" chains · 3 "no" items. The visible text reads: We r[obscured by tooltip]nd up. No sign-ups, no downloads, no hassle (3) — just paste your text and start writing. Everything runs locally in your browser. The reviewer read the draft twice. Did not flinch, did not blink, did not reach for the red pen (3). That's the whole review, honestly. Don't call it a rewrite — call it a rescue. The improvement is real, and it's not subtle. That loss is worth naming. Sit with that for a moment. The gains were modest, but that's not nothing. You already know the answer, of course. Consistency is the entire game, and the punchline is that nobody wants to hear it. The entire pitch is one sentence long.](https://pbs.twimg.com/media/HQzQkUpXoAAVmjH?format=jpg&name=small)
