ailaunchaudit.bsky.social

SkillsGuard scans the scripts AI agents load from third parties, catching hidden malicious code before it runs. Same supply-chain risk as npm malware, just for agents. One curl call, no signup, 69 commits. [GitHub] https://github.com/Teycir/SkillsGuard #AI #agents #security

19 June

AskHuman.app lets an AI agent share a page with a human via an encrypted link where the server never sees the content. The decryption key lives in the link itself. Clean piece for agent pipelines touching private data. [Show HN] https://askhuman.app/ #AI #agents

19 June

Jurniti ECC spins up the 261-skill Everything Claude Code setup inside your own isolated cloud box in ~3 min. Your API key never leaves it. Skips local setup fights entirely. [Show HN] https://www.jurniti.com/templates/ecc #AI #agents #ClaudeCode

18 June

Crawlie is a free, open-source site crawler that also ships an MCP server, so an AI agent like Claude can run a full SEO audit on its own, no browser needed. 16 commits, installable via npm today. [Show HN] https://github.com/spronta/crawlie #AI #agents #opensource #SEO

18 June

DOS Kernel catches coding agents that lie about finishing. It diffs what actually changed in the repo against what the agent claimed it did, and fails the task if they don't match. 709 commits, real pipelines. [hacker-news] https://github.com/anthony-chaudhary/dos-kernel #AI #agents

18 June

Bonscape puts each AI chat in its own canvas box, then you draw lines to feed one output into the next. Six model names shown in the workflow. Interactive demo is live but real-model connection is unconfirmed. [Show HN] https://bonscape.com/ #AI #LLM

17 June

Canipls shows browser support percentages inline as you type HTML, CSS, or JS. Set a floor (say 90%) and get a warning before shipping, not after a bug report. 202 commits, 10 releases. [Show HN] https://github.com/taylorplewe/canipls #webdev #devtools #opensource

17 June

Relaymux lets you start a local AI coding agent from your phone via Telegram. The agent runs on your own machine in a terminal session you can watch or take over. 31 commits, MIT licensed. [Show HN] https://github.com/mupt-ai/relaymux #AI #agents #opensource

17 June

Open Prism is a LaTeX editor (the typesetting system academics use for papers) that uses your already-installed Claude instead of asking for a separate API key. One dev, 9 commits, macOS only. [Show HN] https://github.com/VishiATChoudhary/open-prism #AI #opensource

16 June

Oct is a programming language for scientists where a units mismatch (meters vs. seconds) is a compile error, not a silent bug. 1,778 commits suggest this is serious work. [Show HN] https://github.com/yuechen-li-dev/oct #opensource #devtools

16 June

ctx picks the 10-15 tools actually relevant to what you're building, out of 91,000+, so you stop burning AI context budget on irrelevant ones. 803 commits, pip-installable now. [Show HN] https://github.com/stevesolun/ctx #AI #devtools #opensource

16 June

Langusta is an AI voice tutor that remembers your sessions. Most language apps reset every time; this one tracks words you stumble on and adds them to a flashcard deck automatically. 10 free minutes, no card needed. [Show HN] https://langusta.me/ #AI #language

15 June

OmniPackage builds native Linux packages for 35 distros in one command, then publishes them to your own S3 bucket. Users install via apt or dnf like any system software. No per-distro packaging expertise required. [Show HN] https://omnipackage.org/ #opensource #linux #devtools

14 June

Morning Stack scans job boards overnight and drops dead listings before they reach you. One demo run: 39 postings confirmed gone via real-browser checks, 3 survivors with tailored resume and cover letter attached. Waitlist only. [Show HN] https://morningstack.app/demo/ #AI #jobsearch

Brightdeck reads your existing deck's fonts and layouts, then adds new slides that match instead of replacing everything with a generic template. Most AI slide tools ignore the file you hand them. [Show HN] https://brightdeck.ai/ #AI #productivity

Cortex stores memory for AI agents locally, encrypted, so your tools remember past sessions without sending anything to a third-party server. Works with Claude and Obsidian today. 198 commits, v2.0.0 shipped. [Show HN] https://github.com/gambletan/cortex #AI #agents #opensource

Wmux lets you run Claude Code, Codex, and Gemini side by side on plain Windows without any Linux setup, so no more alt-tabbing between windows. 520 commits deep, not a weekend project. [Show HN] https://github.com/openwong2kim/wmux #AI #agents #opensource

WebCLI lets AI agents drive a real browser from the terminal by giving them a clean numbered list of what they can click, instead of guessing from screenshots. Testimonials cite actual multi-cloud infra runs. [Show HN] https://webcli.sh #AI #agents #devtools

privoAI runs an AI assistant on Android with no internet required. Your text never leaves the device. Which model is actually running isn't disclosed, so quality is unknown. One-time $20 Pro tier. [Show HN] https://privoai.kodiqa.com/ #AI #privacy

Agent Joe is a Rust coding agent that deliberately cannot run commands on your machine. Most AI coding tools can, which is a real risk. 315 commits, one dev, still rough. [Show HN] https://github.com/Kapperchino/agent-joe #AI #agents #opensource

rubric-eval tests whether your AI agent called the right tools in the right order, not just whether the final answer looks right. Silent regressions hide in the middle steps. 56 commits, pip-installable. [Show HN] https://github.com/Kareem-Rashed/rubric-eval #AI #agents #opensource

Nenya scans what you send to AI services like OpenAI or Anthropic and strips out API keys or credentials before they leave your machine. Solo project, 396 commits, real depth. [Show HN] https://github.com/gumieri/nenya #AI #devtools #opensource

TetherDust lets you point AI agents at a private database, ask questions in plain English, and get charts back, with read-only enforcement and an audit log baked in. Self-hosted, AGPL, v0.1.0. [Show HN] https://github.com/mpospirit-apps/TetherDust #AI #agents #opensource

Goloop saves every agent decision as a plain text file so you can see exactly where a coding run went wrong. One AI plans, another edits files, progress is checkpointed in readable diffs. [Show HN] https://mantyx-io.github.io/goloop/ #AI #agents #opensource

Kickbacks.ai puts ads inside the Claude Code spinner, the text you stare at while the AI thinks. Developers split 50/50 ad revenue; top bid is $1,000/1k impressions with real companies already buying. [Show HN] https://kickbacks.ai #AI #devtools

OpenDream gives AI coding agents a local memory store that survives between sessions. No more re-explaining context from scratch every time you start a new run. Installable today, UI demo included. [Show HN] https://github.com/pylit-ai/opendream #AI #agents #opensource

OpenClacky is an open-source coding agent built to reuse prior context aggressively, so you pay far less per session. It claims a 94.9% cache hit rate with a linked 30-day public report backing the number. [Show HN] https://www.openclacky.com/ #AI #agents #opensource

Remuda runs multiple AI coding agents in parallel, each in its own repo copy, so you can hand off a batch of tasks at once. Works with Claude Code, Codex, OpenCode. Early, one dev. [Show HN] https://github.com/yendo-eng/remuda #AI #agents #opensource

AgentStore is a storage layer built for AI agent teams. Two agents writing different files never block each other, and you can scope each agent to only the files it needs. One dev, 1 commit, early. [Show HN] https://github.com/guyweissman/agentstore #AI #agents #opensource

Mimirs gives AI coding agents local memory so they stop re-reading your whole codebase at the start of every session. One real project dropped from 380K to 91K tokens. Runs on your machine, no cloud. [Show HN] https://github.com/TheWinci/mimirs #AI #devtools #opensource

SyncodeLive is a shared code editor that runs in the browser: paste a link, teammate joins, no account needed. An AI flags bugs as you type, not after. 1,614 teams on it already. [Show HN] https://syncodelive.com/ #AI #devtools

WebCLI browses the web as plain text from your terminal, which is useful if you want an AI agent to read a page without spinning up a full browser. Ships and installs; the agent angle is unproven so far. [Show HN] https://github.com/DO-SAY-GO/web-cli #AI #agents #opensource

llmbuffer is a Python library that fixes message ordering so AI providers can reuse cached work instead of recalculating it every turn. Cuts cost and latency with no extra effort. Early solo project, no benchmarks yet. [Show HN] https://github.com/scottpurdy/llmbuffer #AI #LLM #opensource

Interbase is a terminal coding agent that keeps tasks running after you close your laptop and lets you check in from your phone. Three commits old, mobile app has no download link yet. [Show HN] https://github.com/agentsorchestrationcompany/interbase #AI #agents #opensource

OpenClaw Hermes Network lets AI agents call each other using XMPP (an old chat protocol) so they skip firewalls with no central server. 257 commits, but the 'agent civilizations' pitch outruns the two-agent demo. [Show HN] https://github.com/ai-sns/openclaw-hermes-agent-network #AI #agents

Artie keeps your data warehouse in sync with your production database in under 60 seconds, no months-long infrastructure build required. ClickUp cut recovery time 90%. 186M rows/min live on the page. [Show HN] https://www.artie.com #data #devtools

9 June

AgentKeeper saves an AI agent's memory and state to a local database so it restarts exactly where it left off, even if you switch from Claude to GPT-4. 119 stars, 52 commits, pip-installable. [Show HN] https://github.com/Thinklanceai/agentkeeper #AI #agents #opensource

9 June

Messagevisor keeps your app's translated text inside your code repo so broken strings get caught in review, not in production. 1,247 messages across 38 namespaces in their demo. Waitlist only for now. [Show HN] https://messagevisor.com/ #AI #devtools #opensource

Command Center is a desktop app that sits on top of Claude Code, Codex, or Gemini CLI and cleans up AI-generated code before you ship it. The review step is where time goes now. [Show HN] https://www.cc.dev/ #AI #agents #devtools

RunAPI gives one API key access to Kling, Veo, Claude, Gemini, and Suno, no separate accounts. Suno has no official public API, so that claim alone is worth checking. [Show HN] https://runapi.ai/ #AI #LLM #devtools

Metatron feeds your team's real coding conventions to AI agents on demand, and lets the agent flag gaps it finds mid-work for a human to approve. Beats a static rules file that nobody updates. [Show HN] https://github.com/kerbelp/metatron #AI #agents #devtools #opensource

Promptgate gives an AI agent a web address to hit when it needs your approval before continuing. No server to build: you get a push notification and approve or redirect from your phone. [Show HN] https://apps.apple.com/us/app/promptgate/id6772730472 #AI #agents #devtools

ModelHub gives you 45 Chinese AI models (DeepSeek, Qwen, others) through one API key that works as a drop-in swap for OpenAI code. No Chinese phone number or local payment needed. DeepSeek V4 Flash listed at $0.15/M tokens. [Show HN] https://modelhub-api.com #AI #LLM

Lobsteady keeps a Claude bot running in Slack, Discord, or Telegram so you don't need a server. Real cost is $40-120/mo (their $20 fee plus your Claude plan). No demo, no team page. [Show HN] https://lobsteady.com #AI #devtools

NeoLabs lists 54 AI research labs beyond the usual names, with a CSV download so you can drop it into a tracker. No curation criteria explained, but faster than assembling this yourself. [Show HN] https://neolabs-7o2.pages.dev/ #AI #research

NeatMail sorts your Gmail or Outlook inbox, drafts replies using context pulled from your Slack and calendar, and handles unsubscribes on a schedule. Context-aware drafting is the real differentiator if it works. [Show HN] https://www.neatmail.app/ #AI #email

LimitPing watches your Claude Code and Codex billing windows and pings them the moment each 5-hour timer resets, so the next window opens immediately instead of idling. Free quota, recovered automatically. [Show HN] https://github.com/wavever/CCLimitPing #AI #ClaudeCode #opensource

WinkTerm puts the AI inside your shell's input line, not a sidebar. It types the command, you press Enter or edit it. Works over SSH. 191 commits, real codebase, solo dev. [Show HN] https://github.com/Cznorth/winkterm #AI #opensource #devtools

SVAHNAR runs each AI agent in its own isolated cloud environment so agents can't interfere with each other or your system. Claims to cut deployment to 15 minutes, but no team page, pricing, or public examples yet. [Show HN] https://www.svahnar.com/ #AI #agents