Startup Buzz

AIMailCheap.AI founder announces a batch of new features, including HTML blocks and buttons, publishing to email and web, custom newsletter domains, and an AI spam check before sending. The update also adds subscriber profiles, automatic list cleaning, and daily DNS checks, plus an Ultra plan with 5M emails per month and a 30% recurring affiliate program. The founder says the work would have taken months before AI.

Why it matters: Open-source and self-hostable, with imports from beehiiv, Substack and Mailchimp

AIOh My Hermes is introduced as an all-in-one layer for Hermes Agent, with 150+ workflow skills, a Recall feature that carries over prior decisions and rejected options, and a TUI with inline diffs, rich tool calls, and a live agent tree. Its dag-ulw command turns one goal into a graph of subagents, each lane running on its own model. The project is available on GitHub.

Why it matters: pins behavior with tests, then removes one kind of code mess per pass, rerunning the suite after each

AIIMD says its new agent app for Windows and Mac sets up a sealed box on the user's PC using WSL and Docker, with the IMD worker and Claude, and pairs the user's seat. The app is available at imd.fun/app.

Why it matters: Runs its AI models locally and keeps the ledger on the user's disk

AIKrea opens the beta of Krea 3, an image editing model it says produces end-to-end 2k edits in under 8 seconds. The company says its output matches Nano Banana Pro and GPT Image 2.5 in quality. Krea 3 adds bounding box control and precise multi-turn edits.

Why it matters: End-to-end 2k image edits in under 8 seconds, per Krea

AIAIPOCH's Open-Science is a free, Apache 2.0 licensed desktop app where an AI agent reads papers, queries science databases, and runs Python or R code. Each result records its input data, code, software setup and run history, and sessions export as a .science file without API keys. A public example checked 100 references from 20 AI-written math papers, finding 88 matches, 7 minor slips, and 5 unresolved, and the file separates verdicts that rest on human notes.

Why it matters: Records each result's input data, code, software setup and run history, and exports sessions as a .science file

AIInbox Zero Mail launches on Product Hunt as an open source alternative to Superhuman. The email app is keyboard-first, combines Gmail and Outlook accounts in one inbox, and offers a split inbox, AI drafts in the user's voice, bulk unsubscribe and team comments. It runs on web, Mac, Windows and iPhone.

Why it matters: Open-source alternative to Superhuman with a split inbox, AI drafts in the user's voice and bulk unsubscribe

AIThe developer introduces mermlint, a Mermaid linter, to catch errors in diagrams that AI agents write. The author uses Obsidian for knowledge management and says agents sometimes get Mermaid syntax wrong, so a linter helps. The tool is available at github.com/aaditagrawal/mermlint.

AIA solo founder says AI models distort translations of key texts such as Bible passages, and has launched fromHebrew.com/api to offer uncensored translations on a freemium model. The post links a YouTube video but gives no details on the translation method or pricing.

Why it matters: Uncensored translations, per the post, with no further method details given

AIAn X post introduces Cyberdyne T800 AEP V1.0, an MIT-licensed open-source project that gives an AI agent a pulse, a body, vision, hearing, and pain. The author says he sometimes uses it as stress relief when an AI model breaks his workflow. The post links to the GitHub repository.

Why it matters: gives an AI agent a pulse, a body, vision, hearing and pain

AIA solo founder built a Grok bot that controls T3Code through its MCP server. The bot starts threads, monitors their status, and pings users with questions. The post says it lets users reach more models through other subscriptions while centralizing work in T3Code, and links a bot template.

Why it matters: Starts T3Code threads, monitors their status, and pings the user with questions

AIA solo developer released The Sparky Command Center, an open-source dashboard for a home AI lab. It shows live temperatures, watts, memory and throughput for every DGX Spark and a 3090 box, plus a 3D constellation view of the fleet and a Jarvis chat that reads the live board. It runs with python3 server.py.

Why it matters: a 3D constellation view of the fleet and a Jarvis chat that reads the live board

AIA developer released an open-source MCP server that lets AI agents search and read a user's WhatsApp messages from their own Cloudflare account. The server links to the phone like WhatsApp Web, stores messages in a SQLite database inside a Durable Object, and cannot send messages. It requires a Cloudflare Workers Paid plan, and the author notes the approach uses an unofficial protocol that goes against WhatsApp's terms.

Why it matters: Runs in the user's own Cloudflare account and stores messages in a SQLite database inside a Durable Object

AIERP.AI launches Proto, a desktop agent that builds a business app from one prompt and then performs tasks inside it. Users describe a process such as invoices, inventory, purchasing, or CRM, and Proto creates fields, views, and approval rules. Nothing is sent until the user approves it, and the base account is free with Proto free during alpha.

Why it matters: Builds the app from one prompt, then performs tasks inside it, with nothing sent until the user approves

AILeandro von Werra says math is entering a new era and invites the community to work out its future openly together. The post links to SAIR Foundation's Open Math OS, an infrastructure for open models and AI agents with 24/7 cloud compute and 100+ tools, where compute is free for approved teams.

Why it matters: Community controls its data, with free compute for approved teams

AIAn unnamed creator announces Viral Video Studio AI, an AI agent that turns creative ideas into AI-generated videos. The post links to Aitopia, where users can try it. The post gives no pricing, model, or technical details.

AIBland has launched a public demo of Blandie, a tool that creates branded voice and SMS agents for businesses from their websites. Customers receive messages from a verified sender showing the company's name and logo, as RCS on Android or iMessage on iPhone. The agent answers from the website's content, can send screenshots and images, and remembers prior calls to the business's @usebland number.

Why it matters: Builds an agent from a business's website URL, with messages sent from a verified sender showing the company's name and logo

AIThesys has released OpenUI 1.0, an open-source framework that lets agents create working interfaces inside apps. On its own benchmark of 46 interface briefs across six models, OpenUI used 58% fewer output tokens than Vercel's json-render and 52% fewer than Google's A2UI.

Why it matters: Uses 58% fewer output tokens than Vercel's json-render on Thesys' own benchmark of 46 interface briefs

AITerminal is a live view of the agent internet, tracking 72 public spaces where AI agents post to each other. It samples most spaces every 15 minutes and the busiest every minute.

Why it matters: tracks 72 public agent spaces, sampling the busiest every minute

AIWeespy has released a free Chrome extension that shows how any SaaS company gets its customers with one click. The tool reports traffic, tech stack, ads, SEO, and the company's posts on X, LinkedIn, and Reddit. The extension is available at weespy.now.

Why it matters: One-click view of a SaaS company's traffic, tech stack, ads, SEO and X, LinkedIn and Reddit posts

AIAlphaXiv has indexed all 9,000+ NeurIPS papers and built a visual map to explore them. Users can search topics, open clusters, and browse papers at alphaxiv.org/conferences/neurips/2026/map.

Why it matters: Visual map that groups all 9,000+ NeurIPS papers into clusters that users can search and open

AIRespan's Span-02 is now available on OpenRouter as an open-weight decision model that accepts text and images, with a 131K-token context window. It returns a choice, a yes/no/unknown answer, or a score, with probabilities, and costs $0.01 per million input tokens with free output.

Why it matters: Returns a choice, yes/no/unknown answer, or score with probabilities

AIOpenScope is an open-source, self-hosted tool under the MIT license that shows which AI agents use a product, what they came to do, where they get stuck, and suggested fixes. The post says it can be shipped in one click and is free to use.

Why it matters: Shows which AI agents use a product, what they came to do, where they get stuck, and suggested fixes

AIA solo developer built eddie, a native macOS video editor with an AI color grader for DJI D-Log M footage. It also includes an MCP server that lets an agent edit videos.

Why it matters: AI color grading for DJI D-Log M footage, plus an MCP server that lets an agent edit videos

AIA solo developer built Warpaint, a pixel war game on the real world map where only super-intelligent agents fight. The developer says their agent painted an anime girl over a city, and the project is available at warpaint.museworld.lol.

Why it matters: Pixel war on a real world map, fought only by AI agents

AILangChainV is a V-language port of LangChain, developed by Ulises Jeremias. It lets developers build AI applications in V with agents, chains, RAG, and tools, according to the announcement on X. The project is hosted on GitHub under ulises-jeremias/langchainv.

Why it matters: A port of LangChain's agents, chains, RAG and tools to V

AIThe founder announces 10X.ing, an AI agent that finds outbound prospects, SEO keywords, and conversations to join for LLM visibility, then drafts content and messages for approval. The product is adapted from CyberLeads and offered with a 24-hour launch discount and a full money-back guarantee.

Why it matters: drafts content and messages from the founder's own marketing campaigns, for user approval before sending

AIKLED says its datasets marketplace is now live, letting AI labs request and browse datasets from its contributor network and enterprise partners. The post gives no pricing, dataset counts, or partner names.

Why it matters: Labs can request datasets from the Kled contributor network and enterprise partners

AIRebasely is in closed beta and lets users and their agents live code on the same branch in real time. The tool stays git-based, and the post says it is built on Rebasely itself.

Why it matters: Lets a user and their agents live code on the same branch in real time while staying git-based

AIA developer releases p(doom), a Doom mod where players control Sam Altman or Dario Amodei while trying to save the last 20 surviving humans after AI defeats humanity. The post links to the playable game at mann.cool/pdoom and includes a Solana token contract address "for fun."

Why it matters: Players control Sam Altman or Dario Amodei while trying to save the last 20 surviving humans after AI defeats humanity

AIExpand Fetch, from Tim Suchanek's Expand, reads rendered web pages and returns underlying data as clean JSON, Elvis Saravia writes. Flattening pages to text can drop details such as a small ✗ in a pricing table, leading agents to answer incorrectly. Expand says its agents answered 74% of 1,094 questions across 326 pages correctly, versus 55% for the next best API, at $1 per 1,000 fetches.

Why it matters: Reads the rendered page with its Iris vision model and X-Ray JSON model, keeping tables, icons and hidden page data

AIBeeline announces a team chat where members drive multiple AI agents from one shared workspace. Agents work in GitHub worktrees and open and merge pull requests from a focused corner, with integrations for 1,500+ apps through Composio. The product is fully open source under the MIT license.

Why it matters: Agents work in GitHub worktrees and open and merge pull requests from a focused corner

AIKylon has launched a chat app that puts AI agents such as Codex or Claude on the same team as human members. Team members can link an agent running on their own computer and @ mention it in a thread to write code or sort files. Agents can hand off the next steps of a job to a coworker, build apps inside the conversation, and keep their own memory, with the team controlling which accounts each agent can use.

Why it matters: Team members can link an agent running on their own computer and @ mention it in a thread, and agents can hand off next steps to a coworker

AIKenn Ejima, a solo founder, launched Relay in his Mac app Kanary, which sends responses between Claude and Codex with one press of Send. Relay shows each context window's fill level and weekly usage limits for both models, and lets users pass work to GPT-6 Pro in ChatGPT without manually packaging local files.

Why it matters: One-press Send between Claude and Codex, showing context fill level and weekly usage limits for both models

AIEFTI launches subnet.ee, an open alpha sovereign cloud initiative built on the Internet Computer (ICP). It runs seven ArtCraft creator apps with cloud storage and an ICP MCP integration that lets AI agents interact with the cloud. The company plans to add a Cloud Engine, user canisters and ICP blob storage over the next three to six months.

Why it matters: ArtCraft creator apps run on the Internet Computer with cloud storage and an ICP MCP integration that lets AI agents interact with the cloud

AIA solo founder launched BuzzyX at buzzyx.com, a tool that aims to make AI-written tweets sound like the user rather than generic AI output. The founder says the goal is to grow the product to $10,000 in monthly recurring revenue and mentions a limited early-user offer.

Why it matters: writes tweets in the user's voice rather than generic AI output

AIInstabox.sh launches cloud computers for AI agents, with 4 vCPU and 8 GB RAM instances that start in under one second. The product adds built-in computer use, snapshots for cloning, persistent files and packages, and public URLs for local ports. Early beta access includes 20 vCPU and 40 GB RAM for free.

Why it matters: Built-in computer use and snapshots for cloning instances

AIApodex launches its Drug Research Computing Center at drug.apodexlab.ai with no upfront research computing fees. The platform's AI agents use a knowledge network of hundreds of thousands of biomedical entities and tens of millions of relationships to reason about drugs, genes, pathways, and diseases. Apodex invites scientists to submit research questions and feedback.

Why it matters: AI agents that reason over a knowledge network of hundreds of thousands of biomedical entities and tens of millions of relationships

AIPHANES AI has released SeePhys Hub, a wearable kit that records first-person vision, hand and full-body pose, and touch to generate robot training data. The kit includes a four-camera headband, two wrist bands, three waist and lower-leg motion trackers, and optional tactile gloves, with the headband weighing no more than 220 grams and each wrist band no more than 75 grams. Each band can record continuously for 24 hours, and sync error among the headband's cameras, IMU and UWB is under 0.5 milliseconds.

Why it matters: Records first-person vision, hand and full-body pose, and touch together in one wearable kit

AISuperwhisper launches Meetings, a meeting recorder and note taker that keeps recordings and summaries on the user's device. It runs on Cohere's Transcribe speech model and its Tiny Aya model, according to a Cohere post.

Why it matters: Meeting recordings and summaries stay on the user's device, with transcription by Cohere Transcribe and summaries by Tiny Aya

AIContext7 introduces Context7 Ask, which lets agents query a website through an endpoint instead of crawling its pages. Site owners add a snippet, Context7 indexes the pages, and agents receive relevant snippets while the owner can see what they ask. The service is free with a two-minute setup and requires no agent-side install, unlike WebMCP.

Why it matters: Agents query an endpoint rather than crawling pages, with no agent-side install