AI Product Updates: New Features and Releases

Vercel's AI SDK now supports MCP Events, letting agents discover and subscribe to MCP server events while Vercel Connect handles subscription storage, renewals and verified delivery.

bb 0.46.1 adds an MCP server for controlling bb from Claude Code, Grok Bot, Dots and more, plus one-click global CLI install, search settings and inline answers to child threads.

Bland launched a demo where texting a company's website link to the Blandie line builds a branded SMS AI agent in seconds, with iMessage components, Tapbacks and in-chat payments.

Bland released Blandie, which lets users interact with business AI agents over text to get live website answers, book appointments, place orders and pay, while remembering conversations across calls and texts.

Agent Vault keeps agents on placeholder tokens while real card numbers execute inside a TEE, proxies login credentials, and adds spending caps, blocklists and a kill switch.

Agent Vault lets AI agents make payments and log into accounts without seeing card details or passwords, while users set spending limits and approve transactions.

Set up once and agents like Grok Bot, Claude Code and Codex check out with placeholder tokens while real card numbers stay in a secure enclave, with spending caps, blocked sites and a kill switch.

Selfxyz's Agent Vault lets Claude Code, Codex, Grok and other agents share one vault, paying with placeholder tokens instead of the real card number, with spending caps, blocked sites and approvals.

ERP.AI generates ERP, CRM, finance and other business systems from a single prompt, with a desktop agent called Proto to build and operate them; the base account is free and Proto is free during alpha.

Hey Research Lab's remote MCP server is now listed on awesome-remote-mcp-servers and indexed on Glama, letting any AI agent query Robinhood Chain projects with no install or key and get sourced answers.

Self's Agent Vault lets AI agents make payments and log in to websites without exposing card details or passwords, with spending limits and approval rules shared across multiple agents.

NVIDIA's multimodal MoE model Nemotron 3.5 Super VL is available on CoreWeave Serverless Inference, supporting text and image input, reasoning, tool calling and structured outputs.