Nuveki v1.37.0 is live: - Email replies stay in their Gmail thread: the card finds the conversation, you pick Reply in thread or New email - Shipping an update no longer cuts off a reply mid-answer - A workflow waiting on your approval survives a restart https://nuveki.ai

AWS Activate rejected us in under a day. The email gave no reason. The dashboard did: our robots.txt opened with "Disallow: /", which Google reads fine and Python's parser reads as "keep out". We fixed it, went through NVIDIA Inception, and a week later had $10,000 in credits.

AWS Activate rejected us in a day. A week later we had $10,000 in credits.

Nuveki is a chat where a team of AI agents works on your request together. It runs entirely on AWS, and AWS is about 90% of what it costs…

medium.com

AWS can't cap your bill, and its billing data runs 8 to 24 hours late. So we priced Bedrock's live token metrics ourselves, matched September's bill to within six cents, and put three layers of IAM deny under a monthly ceiling. It trips in about five minutes.

AWS can’t cap your bill. So we built a kill switch that trips in five minutes.

AWS billing runs up to a day late. We priced Bedrock's token metrics live, matched the bill within six cents, and capped spend with IAM.

medium.com

Nuveki v1.35.0 is live: - Three rewriters make text read like a person wrote it, in your voice or a native speaker's - Emails drawn as emails: fill the placeholders, open in Gmail or Outlook - The app downloads about half as much before it shows anything nuveki.ai

Nuveki v1.34.0 is live: - Gemini 4 Argon replaces Gemini 3.1 Pro: about a million tokens in, up to 65,536 out - Qwen3.8 Max is pinned to its 0902 version, so answers hold week to week - One agent answers, unless a second makes the answer better https://nuveki.ai

Three things broke first when our AI agents started working as a team: success claims with no tool call behind them, plans cut off at max_tokens while the p90 looked fine, and steps that never saw the work they depended on. What we check now: https://medium.com/p/1bea43b0de71

Ask Nuveki's PowerPoint agent for a deck in one sentence: five slides, one in two columns. It writes a real .pptx to your Library, and the file now opens as slides right in the chat: filmstrip, all-slides grid, no download. New in v1.32, live today. 98-second demo: https://youtu.be/9wHY-nCo0aQ

Most AI chats answer in text. One Nuveki answer can hold a formula that plots itself with sliders, a molecule you open into its structure, a live map, a zoomable diagram and a tune as real sheet music. Nothing to install. 99s demo: youtu.be/1yd8UpeF21U

Should a 20-person agency try a four-day week? One message: research first, then operations, people and legal read the same facts in parallel, then one agent weighs them and names the biggest disagreement. Nuveki turns that sentence into the run. Demo: youtu.be/mcPP3WK3Sks

Our planner's p90 was 5,294 tokens under an 8,192 cap. Looked safe. Then we counted: 19 of 91 plans ended at exactly 8,192, cut off mid-JSON. A spike on the cap is censored data. If you read token percentiles under a max_tokens, count the hits on the limit first.

Multi-agent AI shouldn't be a black box. One message, three agents in order: research, then an offer, then a critic. The run strip shows who's running and who's waiting. One click opens the Inspector: the graph, each agent's exact brief, every tool call. 86s demo: youtu.be/6-ecMcSLCPU

Nuveki v1.29.0 is live: - Claude Sonnet 5.5 is the default. 107 more agents moved to it and answer about twice as fast, at the same or a lower cost - IFC building models open in 3D: sections, build order, sun study - CoinGecko and Wolfram, no setup nuveki.ai