Highlights
What happened at the companies we cover, by day, newest first. Every item links to its sources.
Showing 1–20 of 28 highlights
Thursday, Nov 5 28 stories
Lovable spotlights Parallel connector for live web search in apps
LaunchX
On November 5 Lovable’s official account promoted Parallel in Lovable so builders can add live web search or deep research into apps from the connectors menu or by tagging it in chat. Engagement was strong on the announcement thread, and Lovable also posted a product roundup video covering growth tools, free chat, WhatsApp, more in-app AI models, and deploy-anywhere options. Parallel itself appeared earlier in the connector catalog; this was a major visibility push.
Sources
Baseten Sandboxes: microVM agent computers with 25 ms resume, free until Nov 16
LaunchNewsOther
On November 5 Baseten launched Sandboxes as the first piece of an agent cloud: isolated microVMs with instant suspend/resume that snapshot filesystem, memory, and processes so agents keep state without paying for idle compute. Boot from standby is about 25 ms, with roadmap notes for Agent Drive storage and Agent Proxy networking. Sandboxes are free to run until November 16, 2026.
Sources
Hugging Face details Phonon on-device TTS after strong leaderboard results
NewsNews
A Hugging Face blog walked through Phonon, the small on-device open TTS stack that topped several Open TTS Leaderboard intelligibility and CPU streaming checks. Phonon ships ~65–75 MB models for five languages, with Phonon 31m under Apache 2.0, Transformers integration, and a dedicated ptts inference framework across CPU, GPU, and NPU backends. The post reports more than 200,000 downloads in the first two weeks.
Sources
Hugging Face ships ML Intern as an installable Slack bot
LaunchNewsX
Hugging Face made its open-source ML Intern agent available as a Slack bot that evaluates models, builds datasets, and fine-tunes under each user's own Hub account. Threads start with a zero compute budget so Jobs and sandboxes need explicit approval, and users can bring MCP servers from HuggingChat. The bot is part of Chat UI and can also be self-hosted.
Sources
LeRobot v0.7.0 adds interactive robot prompting and disk-less training
LaunchNewsX
Hugging Face released LeRobot v0.7.0 with interactive rollouts so operators can change subtasks, ask visual questions, and autosteer a running robot without restarting. The release streams Hub-scale datasets without full downloads, adds remote policy servers for light robots, and ships new world action models and VLAs including Cosmos 3 and FLUX 3 Action. Training now targets multi-billion-parameter Physical AI workloads with FSDP2 and distributed checkpoints.
Sources
Vercel Schedules enters public beta for apps and agents
LaunchNewsX
Vercel opened Schedules in public beta so apps and agents can define static jobs in project config and create dynamic one-time or recurring schedules at runtime, including on Preview. Jobs can target Functions or Queue topics and succeed Cron Jobs while leaving existing cron working. Official X posts drove heavy engagement, and community explainers followed the same day. Gemini Omni 1.2 Flash also landed on AI Gateway for video generation alongside the Schedules ship.
Sources
KDDI Biz Edge will sell Genspark to Japan SMEs
NewsNewsX
Alongside TeamBrain, Genspark signed a strategic partnership with KDDI Biz Edge to sell and support Genspark for Japanese mid-sized and small businesses. KDDI Biz Edge will handle sales through go-live and adoption support, with service sales planned from January 2027 under a Mobile Work AI pitch for smartphone-first field work. CEO comments from both sides framed the deal as bringing agent workflows to industries where desktop AI never stuck. Multiple Japanese outlets covered the agreement the same day as the TeamBrain launch.
Sources
Genspark ships TeamBrain shared context for agents
LaunchNewsXBluesky
Genspark launched TeamBrain for enterprise customers, an organizational knowledge layer that compiles documents, email, chat, and SaaS data into shared context both people and Genspark agents can use. The product auto-structures material without heavy upfront tagging and is meant to sit beside SecondBrain, AgentBase, and GenTeam. The same announcement tied TeamBrain to a Japan go-to-market push so mid-market teams can adopt agent workflows without a large data-engineering program. Coverage across Japanese tech outlets amplified the launch the same day.
Sources
Cursor lets cloud agents build and deploy live internal apps
LaunchNews
Cursor shipped Build Live Apps so a cloud agent can scaffold a web app, provision Postgres and SSO auth, run security checks, and publish a shareable link behind Cursor sign-in. Team admins can review apps and secrets, add review rules, and unpublish. The feature targets internal tools across ops, sales, and eng rather than another IDE model upgrade.
Sources
Higgsfield adds Gemini Omni Flash 1.2 and GPT Image 2.6 Flare/Sunburst
LaunchOtherNewsX
On November 5 Higgsfield brought Google Gemini Omni Flash 1.2 for video with native audio, 3–15 second clips up to 4K, references, frames, and extend modes, plus OpenAI GPT Image 2.6 in Flare and Sunburst variants with GIF, SVG, and transparent-background outputs. Combined with November 4’s Kling 4.0 Flash for cheaper short tests, the catalog update is aimed at creators iterating faster across image and video.
Sources
Business Insider: Harvey held early IPO talks with Latham & Watkins counsel
NewsNews
On November 5 Business Insider reported Harvey has held preliminary IPO discussions with lawyers at Latham & Watkins after a fall $550M raise and a stated $400M ARR run rate, with no listing imminent. The piece notes executive hires with public-company experience and a $15.5B valuation backdrop. Harvey declined to comment. This is exploratory market-prep news, not an S-1 filing.
Sources
FORUM Chicago: Harvey Graph, Outlook agents, in-house Intake, and Fact Graph debut
LaunchNewsOther
At FORUM Chicago on November 5 Harvey previewed Harvey Graph for shared matter context across agents, a two-way iManage sync in early access that preserves ethical walls, Outlook agents for triage and drafting with human approval, in-house Intake that routes contract requests from email, and a litigation Fact Graph linking facts to sources. Several capabilities are early access or waitlist rather than broad GA. CEO Winston Weinberg framed 2027 as moving from prompted tools to governed agent teams.
Sources
LMSYS highlights Decagon talk on RL for production agents
DiscussionX
LMSYS Org promoted a Decagon talk by Cyrus Asgari on reinforcement learning for a production customer-facing agent at hundreds of billions of parameters, covering synthetic data, reward design, and training-inference mismatch. The appearance keeps Decagon visible in the open research community around agent reliability rather than announcing a new product. Separately, hiring posts listed Decagon among NYC startups recruiting senior engineers.
Sources
Coverage amplifies Suno MIDI Create and Studio pitch tools
NewsNewsX
Producer-focused coverage on November 5 walked through Suno’s new MIDI Create uploads and Studio’s finer pitch and tempo-aware import controls. Writers positioned MIDI as the path for composers who need exact note fidelity rather than prompt-only generation. Community posts on X celebrated the launch while some Premier subscribers continued to complain about credit burn and pricing. No new company blog beyond the November 4 MIDI post appeared that day.
Sources
ElevenLabs lets ElevenMusic users publish straight to SoundCloud
NewsNewsXBluesky
On November 5, ElevenLabs and SoundCloud announced that creators can publish AI-generated tracks from ElevenMusic directly to SoundCloud without manual file uploads. SoundCloud Artist Pro users get three months of ElevenMusic Pro, and ElevenLabs said users have already created more than 25 million tracks on its music models. The move sits against platforms that label or ban AI music, making distribution policy as important as model quality for builders shipping AI audio.
Sources
EliseAI tests OpenAI GPT-Live-1 for housing and healthcare voice
NewsNews
Reports say EliseAI is testing OpenAI’s GPT-Live-1 for real-time resident and patient voice support, focusing on interruptions, latency, reliability, and handoffs to humans. Leadership scheduled a November 17 San Francisco event with OpenAI to discuss production testing in housing and healthcare, while EliseAI also plans to show its Apollo team agent at OPTECH 2026 mid-November.
Sources
Parloa Lens expands to evaluate AI agents on any stack plus human contact-center agents
LaunchNewsXOther
On November 5 Parloa expanded Lens into a single observability layer for AI agents built outside Parloa and for human agents in any contact center. Teams can stream or batch transcripts, optionally add tool traces and prompts/SOPs, and apply shared performance, quality, and compliance standards. Lens aggregates human results alongside AI metrics rather than using AI to score individual human performance. The move targets fragmented CX measurement when customers bounce between bots and people.
Sources
Runway publishes real-time 57B Characters serving research
NewsNews
Runway Research published how it serves a roughly 57B-parameter autoregressive video diffusion transformer for live Characters conversations at 24 fps. Optimizations cut a four-H100 step from about 241ms to 159ms, with a fused GEMM/all-reduce path at 147ms, and production telemetry showed most sessions sustaining above 23.5 fps. The paper is a systems proof point for batch-size-one interactive video, not a new consumer model launch.
Sources
Runway Dev adds Kling 4.0 Flash, Gemini Omni Flash, GPT Image 2.6
LaunchNews
Runway Dev's November 5 changelog added Kling 4.0 Flash for faster 720p video with audio, Gemini Omni 1.2 Flash across 360p to 4K video modes, and GPT Image 2.6 Flare and Sunburst for high-resolution image generation with large reference budgets. Pricing is credit-based with separate input-video duration rules for Kling Flash. The drop keeps Runway Dev positioned as a multi-lab media API rather than a single-model stack.
Sources
Microsoft declines option to expand Texas deal with Nscale
NewsNewsXHNBluesky
Reuters and Bloomberg reported that Microsoft will not exercise an option to more than double its Texas computing agreement with Nscale, after power-connection and delivery issues across neocloud partners. Sources said the decision was made months earlier and that the existing Texas contract remains intact, with Nscale’s disclosed ~$103B contract total unchanged as it seeks other customers for spare capacity. Bloomberg framed Microsoft as souring on neoclouds including IREN, Nebius, and Nscale amid delays and costs, while shifting emphasis to its own data centers. The story hit hard as neocloud equities sold off and traders debated whether optional phases were ever solid backlog.
Sources