📡 radar
New launches worth stealing ideas from
Models, agents, tiny tools — the ships we'd actually board. Curated by hand, updated whenever something good launches. No sponsored slots, ever.
🗞 The Shipping News
The best of this feed plus our build log, once a week.
signups run on manifest · one of our seven tools · read issue #1 →
- securityJul 20
A filterable directory of AI agent sandboxes and jails pleasedonotescape.com
If you let an agent run code, you have to decide what it is allowed to touch. This collects the sandboxing and isolation options in one filterable place instead of scattered across READMEs.
- agentsJul 20
Creed — one portable context file that follows you across every agent producthunt.com
A single file describing who you are and how you work, meant to be handed to any agent so you stop re-explaining your preferences. Worth a look if you run more than one coding assistant.
- infraJul 19
BaseRT — a local inference runtime that claims ~6x faster decoding than llama.cpp producthunt.com
A new runtime for running models locally, benchmarked at 6.4x faster than llama.cpp and 3.9x faster than MLX. If the numbers hold on your hardware, that is the kind of speedup that changes what you can run on a laptop.
- voiceJul 19
OfflineTTS — text-to-speech and speech-to-text that runs fully in the browser, locally offlinetts.com
Free TTS and STT that run entirely in the browser with no server round-trip, so audio never leaves the device. Handy when you want voice features without a per-call API bill or a privacy review.
- marketingJul 19
OpenSEO — an open-source alternative to Ahrefs for tracking your own SEO producthunt.com
An open-source take on Ahrefs: backlinks, keywords, and site audits you can self-host instead of renting. Worth a look if SEO tooling costs more than your whole indie stack.
- securityJul 18
Honeypot Live — watch bots attack an SSH honeypot in real time honeypotlive.cc
A live feed of automated bots probing and hammering a public SSH honeypot, streamed as it happens. A quick, visceral reminder of what hits any server the moment you expose it to the internet.
- devtoolsJul 17
Leaves — a text-UI treemap for seeing where your disk space actually went github.com
A terminal treemap that shows what is eating your disk, so you can see the shape of the problem instead of scrolling du output. Handy the next time a build cache quietly swallows 40GB.
- devtoolsJul 17
Mojibake — a small, low-level Unicode library written in C mojibake.zaerl.com
A tiny, dependency-light Unicode library in C for when you need to handle codepoints correctly without pulling in something as heavy as ICU. Low-level by design, if that is the layer you are working at.
- agentsJul 17
DoorDash ships dd-cli — order food, or let an agent do it, from the terminal techcrunch.com
DoorDash opened a beta of dd-cli, a command-line tool that lets you or an AI agent search stores, build a cart, and place an order from the terminal. A small but real step toward agents handling everyday commerce.
- infraJul 16
Firefox now runs entirely in WebAssembly, inside a browser tab developer.puter.com
Puter compiled the whole browser to WASM — no server, no video streaming, just a tab. A useful reminder that "too big to port" keeps shrinking as a category.
- devtoolsJul 16
StyleSeed — a design-rules engine so AI agents stop shipping generic UI github.com
You write your design system as machine-readable rules and your coding agent follows them, instead of defaulting to the same gradient landing page every time. We feel mildly called out, and we are trying it.
- agentsJul 16
Nitrosend gives AI agents their own email accounts producthunt.com
Agents sign up for services, send, and reply without borrowing your inbox. Email is still the universal API, so agent-native email was inevitable — the interesting part will be how verification and abuse handling shake out.
- modelsJul 15
Thinking Machines releases Inkling, its first open model techcrunch.com
The lab's first open model, built for people who'd rather fine-tune a specialist than rent a one-size-fits-all API. Open weights mean you can actually run it yourself.
- devtoolsJul 15
ccshare — multiplayer Claude Code for shared agent sessions producthunt.com
Share a live Claude Code session so two people, or a whole team, can drive the same agent together. Pair programming, but for the thing writing your code.
- infraJul 15
misa77 — a compression codec that decodes ~2x faster than LZ4 github.com
A new codec that decodes about 2x faster than LZ4 while landing better compression ratios. Worth a look if you're shuffling a lot of bytes on a hot path.
- modelsJul 9
OpenAI ships the GPT-5.6 family (Sol, Terra, Luna) techcrunch.com
Three tiers: Sol (flagship, pitched as their best coding model yet), Terra (everyday), Luna (cheap). For tool builders the headline is unit economics — fewer output tokens at a lower price changes what's viable to automate.
- devtoolsJul 9
Aura — open-source IDE for controlling AI coding agents producthunt.com
Agents + git + intent as first-class concepts, with built-in loops. The interesting bet: the IDE of the agent era looks more like a control room than a text editor.
- infraJul 9
Auriko — a trading desk for LLM calls producthunt.com
Treats model usage like an order book. Token costs are quietly becoming a first-class engineering concern, and tooling is showing up to match.
- agentsJul 9
Timbal AI — agents, workflows, and apps in one stack producthunt.com
Another run at the all-in-one agent stack. The pitch keeps getting made; the question to ask is always the escape hatch when you outgrow it.
- securityJul 9
Perfai — security scanning for vibe-coded apps producthunt.com
Finds and fixes live vulnerabilities in AI-built apps with one prompt. A whole security category is forming around code nobody read before shipping.
- voiceJul 9
Lispr — hold a key, speak, it types anywhere producthunt.com
System-wide voice input as a single-purpose tool. Tiny things that do one job well are having a moment (we're biased).
- voiceJul 9
GPT-Live — full-duplex voice for ChatGPT producthunt.com
Talking over each other finally works — voice AI stops feeling like a walkie-talkie. Full-duplex is the difference between a demo and a conversation.
- infraJul 9
Opper AI — a European AI gateway for agents producthunt.com
Routing, compliance, and data locality for teams that can't send everything to us-east-1. Region-bound AI infra is turning into its own market.
Traffic on this page is tracked with our own Pulse — numbers public on /open.