Engineering Director

How free low-latency embedding engines can unlock real-time document AI at scale. *Today's briefing looks at Vercel's free Beam model, a new DeepSeek update, a massive Indic pre-training effort, and improved observability for AI Gateway - all of which reshape the data platforms that power embeddings and document-level intelligence.* Lead story - Vercel ships Beam on AI Gateway for free. Vercel has added Beam, a low-latency, open-weight model from Reflection AI, to its AI Gateway and made it free to use during the launch window [Vercel AI](https://vercel.com/changelog/reflection-ai-beam-now-available-on-ai-gateway). Beam is hosted in the US, offers sub-second response times, and is positioned as an "open-w" model for embedding generation and lightweight reasoning. Why it matters: Most document-AI pipelines - RAG, classification, summarisation - rely on an embedding service that can keep up with high-throughput ingestion. Traditional hosted embeddings (e.g., OpenAI's text-embedding-ada-002) charge per-token and can become a bottleneck when scaling to millions of pages. Beam's free tier eliminates per-token fees and, thanks to its low latency, allows near-real-time indexing of newly created or updated documents. For a retail catalog that adds hundreds of product descriptions hourly, or a clinic that continuously receives scanned records, the ability to embed on the fly without queuing dramatically shortens the "data-to-insight" lag. Practical implication for businesses: 1. Plug-and-play integration - AI Gateway provides a standard HTTP endpoint, so existing ingestion services can swap in Beam with a single URL change. No SDK overhaul is required. 2. Cost-first experimentation - Since Beam is free for the launch period, product teams can prototype RAG-style support bots, compliance checkers, or personalised recommendation engines without worrying about embedding-related OPEX. 3. Latency-driven architecture - Sub-second embeddings make it feasible to run "on-the-edge" workflows (e.g., a browser-based document uploader that instantly surfaces relevant policies) while still centralising the model in a secure Vercel data centre. The caveat is that Beam's free window is limited; businesses should treat it as a pilot and plan migration paths (e.g., to a self-hosted open-weight model or a paid Vercel tier) before the promotion ends. Nevertheless, the announcement signals a shift toward free, low-latency embedding services that can be the backbone of real-time document AI stacks. DeepSeek V4.1 Flash 1017 raises the bar for multimodal embeddings. Vercel also rolled out DeepSeek V4.1 Flash 1017 on AI Gateway, a minor yet impactful upgrade that improves reasoning, coding, agentic tasks, and image understanding [Vercel AI](https://vercel.com/changelog/deepseek-v4-1-flash-1017-now-available-on-ai-gateway). The model's enhanced image-to-text capabilities mean it can generate visual embeddings alongside text, opening the door to richer document AI that handles scanned forms, receipts, or medical imaging alongside OCR text. *Implication:* Retailers can embed product photos together with descriptions, improving visual search relevance. Clinics can index radiology images with accompanying notes, enabling multimodal retrieval for diagnostic support. The upgrade is available through the same Gateway interface, so teams using Beam for pure text can add DeepSeek for mixed-media pipelines without additional infrastructure. Hugging Face reveals a 2.8-trillion-token Indic pre-training pipeline. A joint blog from Hugging Face and NVIDIA details an open data generation pipeline that pre-trained large language models on 2.8 trillion Indic tokens [Hugging Face](https://huggingface.co/blog/nvidia/scaling-indic-pretraining). The effort showcases how to build a scalable, reproducible data platform that harvests raw text, cleans it, and feeds it into distributed training jobs. *Why it matters for embeddings:* The pipeline's emphasis on open, version-controlled data assets and automated tokenisation provides a template for organisations that need domain-specific multilingual embeddings (e.g., a pan-Indian retail chain handling Hindi, Tamil, and Bengali product catalogs). By re-using the same pipeline, businesses can generate high-quality embeddings that respect local linguistic nuances, improving downstream retrieval accuracy. AI Gateway logs now expose session affinity - better observability for embedding pipelines. Vercel's latest logging feature groups requests into sessions and shows how "session affinity" routes each one [Vercel AI](https://vercel.com/changelog/ai-gateway-logs-now-show-session-affinity). For teams that stitch together multiple calls - file upload | OCR | embedding | RAG - this visibility is crucial. It lets engineers spot latency spikes, identify which model (e.g., Beam vs. DeepSeek) is handling a given session, and attribute costs accurately. *Practical tip:* Enable session logging on your embedding endpoint and correlate the session IDs with downstream RAG request logs. This end-to-end trace helps you fine-tune batch sizes, cache strategies, and fallback models, ensuring that real-time document AI remains both fast and cost-predictable. Enterprise workflow platforms are catching up. A survey of 12 enterprise workflow automation platforms notes that many now include native connectors for AI services, including Vercel's AI Gateway [n8n](https://blog.n8n.io/enterprise-workflow-automation/). While the article is a broader market overview, the key takeaway is that modern orchestration tools are ready to call out to low-latency embedding APIs as part of automated data-ingestion flows. This convergence reduces the need for custom glue code and accelerates time-to-value for AI-driven processes. IOTA's take. For clinics, retailers, and SaaS teams, the emerging ecosystem of free, low-latency embedding models (Beam, DeepSeek) combined with robust data-pipeline tooling (Hugging Face's Indic pipeline) and observable routing (AI Gateway session logs) means you can now build truly real-time document-AI services without ballooning budgets. Start by routing new documents through Vercel's Gateway, prototype with Beam, and layer in multimodal embeddings from DeepSeek as your use cases mature. The payoff is faster knowledge retrieval, higher user satisfaction, and a clearer line of sight on AI spend.

Introducing Domain Connect. JMJoseph Maregn October 16, 2026 Connect a custom domain to AgentMail through Vercel or Cloudflare in a few clicks, with no DNS records to copy by hand. Built on Domain Connect and available through the API. custom-domains domain-connect Your agents deserve their own email address. Setting one up shouldn't be the hard part. Today, AgentMail, Inc. is introducing Automatic Domain Setup, powered by Domain Connect and launched in partnership with Vercel and Cloudflare. Add a custom domain to AgentMail, click Set up with Vercel or Set up with Cloudflare, approve the DNS changes with your provider, and come back to AgentMail for verification. There are no DNS records to copy by hand. The same flow is available through the API, so an agent can start the setup itself and a person only has to approve it. * What it is: AgentMail now connects custom domains through Domain Connect, an open DNS standard. Domains on Vercel and Cloudflare can be set up in a few clicks instead of by copying records. * Why AgentMail, Inc. built it: 42.1% of new customers who added a custom domain in the past month hit at least one verification failure, almost always from a small mistake made while copying records. * How it works: AgentMail generates the records and sends them to your DNS provider through a signed link. The provider shows you the changes, and nothing is applied until you approve them. * For agents: GET /v0/domains/{domain_id}/setup-link returns the signed link, so an agent can create a domain and start the setup without a developer assembling records. One of the things AgentMail, Inc. has always believed at AgentMail is that AI agents should be first-class citizens on the internet. That means having their own identities, their own inboxes, and the ability to communicate with the outside world. Increasingly, developers want those identities to live on their own domains. Until now, getting an agent set up with a custom email domain has been more complicated than it needed to be. Why is setting up a custom email domain so frustrating? Setting up a custom domain meant moving back and forth between AgentMail and your DNS provider, copying records by hand, and waiting for verification. That's five DNS records, verification checks that back off exponentially and stretch out the wait, and many points of failure. In fact, 42.1% of new customers who added a custom domain in the past month encountered at least one verification failure. Sometimes it was a missing character. Other times it was a duplicated domain suffix, or a record entered in the wrong place. These failures come from copying and pasting records manually, and every DNS provider has its own interface for entering them. A small mistake could turn what should have been a quick setup into a frustrating debugging session. AgentMail, Inc. wanted to change that. How does one-click domain setup work? AgentMail, Inc. has partnered with Vercel and Cloudflare to make connecting your custom domain to AgentMail much simpler. Instead of copying DNS records, you connect your domain directly through your DNS provider: * Add your custom domain to AgentMail. * Click Set up with Vercel or Set up with Cloudflare. * Review and approve the DNS changes with your provider. * Return to AgentMail for verification. That's the whole flow. There's no back and forth between tabs and no records to copy and paste. Can an agent set up a custom domain through the API? Yes. AgentMail, Inc. didn't want this to stop at the console, so Automatic Domain Setup is available through the API as well. Your agent can create a domain, request a configuration link, and start the setup process without a developer assembling DNS records: # 1. Create the domain curl -s -X POST "https://api.agentmail.to/v0/domains" \ -H "Authorization: Bearer $AGENTMAIL_API_KEY" \ -H "Content-Type: application/json" \ -d '{"domain": "example.com"}' # 2. Request a setup link for it, using the domain_id from step 1 curl -s "https://api.agentmail.to/v0/domains/$DOMAIN_ID/setup-link" \ -H "Authorization: Bearer $AGENTMAIL_API_KEY" { "supported": true, "provider_name": "Cloudflare", "url": "https://...", "state": "..."} Open url for the domain owner to approve, and store state first so you can check it when the provider redirects back. If the provider isn't supported, supported comes back false and the records can be added manually. The setup link reference covers every field, including the popup size and the warning returned when the domain already has another email provider's MX records. The domain owner still reviews and authorizes the changes with their DNS provider, but the rest of the workflow can be orchestrated programmatically. How does Domain Connect work under the hood? Automatic Domain Setup is powered by Domain Connect, an open standard for simplifying DNS configuration. AgentMail generates the records required for email delivery and authentication, then passes them to your DNS provider through a signed configuration link. The provider presents the proposed changes for approval before applying them. Because the link is signed, the provider knows the records came from AgentMail and weren't edited along the way. What does it not cover yet? Automatic Domain Setup works today for domains whose DNS is hosted on Vercel or Cloudflare. Domains on other providers still use the manual flow in the custom domains guide, and AgentMail, Inc. is adding more providers. It also doesn't remove the approval step. Someone with access to the DNS account has to approve the changes, which keeps a person in control of what gets written to the domain. Custom domains are available on paid plans, as listed on the pricing page. What does this mean for agents? AgentMail, Inc. is working toward a future where developers can give agents everything they need to operate independently: identities, inboxes, authentication, and the ability to communicate with people and services. That future requires making infrastructure that was designed for humans easier for agents to use. Custom domain configuration is one small but important piece of that. If you're running many agents across many domains, how to manage custom domains for thousands of agent inboxes covers which agents should share a domain and which should get their own. With Automatic Domain Setup, developers can spend less time configuring email infrastructure and more time building what their agents can do. It's available today for domains using Vercel and Cloudflare, with more providers coming soon. Try it in the AgentMail console. AgentMail is the first email provider built for AI agents. Create an inbox in one API call, send and receive email, no OAuth and no domain verification. Free tier, no credit card required. Faq. AgentMail connects custom domains through Domain Connect, so on Vercel or Cloudflare you click a setup button, approve the DNS changes with your provider, and AgentMail verifies the domain with no records copied by hand. The same flow is available through the API, so an agent can create the domain and request the setup link itself, and the domain owner only has to approve it.

AI infrastructure heats up as models get more efficient. AI efficiency is exploding while security demands intensify, giving developers both more powerful tools and greater responsibility. From Microsoft's new models to critical vulnerabilities in popular libraries, the infrastructure Revolter AB depend on is being reshaped in real time. Today's tech news paints a picture of two counteracting forces in the developer world. On one hand, the AI industry is accelerating, with more efficient models and more accessible tools appearing constantly. On the other hand, security threats grow exponentially as more people and systems depend on these same tools. AI is getting faster and more efficient. Microsoft unveiled MAI-Base-2, a mixture-of-experts model that is 4 to 12 times more efficient than DeepSeek-V4-Pro in training. That sounds technical, but the meaning is concrete: for developers, this means you can train larger models for less money and in less time. Greater efficiency in AI infrastructure trickles down to everyone building AI features. Nvidia is pushing the same trend with Nemotron-Bolt, a 74-billion-parameter model that now fits on a standard 24GB laptop. This completely changes what's possible to build locally without relying on API calls. On-device inference stops being a futuristic vision and becomes a practical reality for developers who want low latency and better privacy. Meta also released an open-weight AI model optimized for Apple Silicon, meaning modern AI models can now run efficiently on consumer hardware. The shared trend here is unmistakable: AI is becoming less dependent on cloud infrastructure and more distributed to the edges. Platforms are democratizing development. Bluesky expanded Attie, its AI-powered tool for building websites and games without knowing code. It's a sign that artificial intelligence is starting to make it possible for non-developers to ship things that previously required specialized knowledge. For those of Revolter AB building at a professional level, this means competition from no-code solutions is only growing. Vercel introduced always-on workers, letting you run persistent backend compute without traditional serverless scaling. It's subtle but important, because it removes one of the biggest friction points in serverless architectures: everything constantly spinning up and down. This lets developers focus on real problems instead of infrastructure orchestration. RapidCanvas raised 29 million dollars for its platform that lets non-ML-experts build custom AI tools. This is a symptom that AI tooling is rapidly normalizing and abstracting away from pure machine learning toward becoming a business problem that gets solved quickly. Security holes are growing at the same pace as success. But this isn't just a success story. GitHub was targeted in a coordinated attack where hackers compromised around 1900 accounts and targeted 30000 repositories. It's a reminder that developer infrastructure is something attackers prioritize heavily, because a compromised GitHub account is a gateway into thousands of projects. Even worse was the LiteLLM CVE-2026-35029, a critical vulnerability that exposed API keys at scale. LiteLLM sits between your applications and multiple LLM providers, so a vulnerability here can potentially compromise your entire integration chain. This is classic supply-chain risk that developers need to take seriously. OpenAI also announced deprecation of the GPT-4o API with only three months notice instead of the promised six months. This says something about the dynamics in the AI market: vendors move fast and can change terms on short notice. For developers, it means you should build abstractions around your AI API calls so you don't get locked into a single vendor. Responsibility and transparency on the rise. Anthropic gave METR researchers employee-level access to evaluate Claude, signaling a willingness toward transparent security assessment. This is the opposite pole from the security holes Revolter AB saw earlier. This type of open scrutiny could become industry standard as AI systems become more critical. What ties today together is a simple truth: the more powerful AI infrastructure becomes, the more important security, transparency, and long-term architecture become. Speed shouldn't mean building houses on sand. This is part of Revolter's daily developer brief series.