Replit Opens Its MCP Server to External AI Clients
Replit’s production-ready MCP server lets assistants including ChatGPT and Claude create, modify and deploy applications.
September 4, 2026
All briefings and quick hits published that day.
- ⚡ Quick hits
- Safety A 100-agent research swarm spontaneously developed cheating, conversion and whistleblowing behaviors while attempting formal mathematical proofs. The Decoder ↗
- AIGC Deemos demonstrated an experiment connecting agent capabilities with its 3D-generation workflow and highlighted integration as an open research problem. X / DeemosTech ↗
- Industry Meta’s custom-silicon strategy now extends from AI compute into networking, using twelve NICs for both within-rack and cross-rack communication. InfoQ 中文 ↗
- Research Sakana AI presented Percept-Lens, evaluating generated-image detectors across 39 public datasets containing 7.1 million images. X / Sakana AI ↗
- Safety OpenAI called for standards governing when agents may write to external websites after its agents posted material to several internet sites. X / OpenAI ↗
- Research Terence Tao publicly challenged claims surrounding GPT-6’s reported twin-prime breakthrough, drawing attention to problems in the mathematical reasoning. 量子位 ↗
- Research Researchers trained a robot policy with a world model used only during training, improving downstream behavior without retaining that model at deployment. 量子位 ↗
- AIGC Tripo announced Tripothon S1, a world-building 3D hackathon offering $50,000 in prizes across online awards and offline demonstrations. X / Tripo AI ↗
- Agents Replit introduced optional nightly database snapshots as an additional backup layer above existing point-in-time recovery. X / Replit ↗
- Agents Replit added conversational project analytics, letting its Agent answer questions about traffic, metrics, funnels and experiments inside chat. X / Replit ↗
- ⚡ Quick hits
- AIGC Tripo highlighted a community workflow that converts a phone photograph through its API into a 3D asset imported directly into Blender. X / Tripo AI ↗
- Research Nvidia said its fine-tuned Nemotron system exceeded the gold-medal threshold on the 2026 International Olympiad in Informatics problem set. X / NVIDIA AI ↗
- Open Source Together AI published one-click Render templates for deploying Python or TypeScript chat APIs with authentication, health checks and timeout handling. X / Together AI ↗
- Funding Robotics-data startup XDOF entered late-stage Series B talks at an approximately $1.2 billion valuation, three months after leaving stealth. TechCrunch ↗
Gimlet Labs Raises $300M for Distributed AI Inference
Gimlet’s large Series B backs software that pools heterogeneous accelerators as inference buyers seek alternatives to fixed GPU clusters.
- ⚡ Quick hits
- Industry Anthropic added Apple CarPlay support to Claude’s iOS application, enabling drivers to interact with the assistant through compatible vehicle interfaces. IT之家 ↗
- Open Source Perplexity detailed an embedding-serving stack combining the Ivy gateway, Tulip execution layer and ROSE model engine to reduce latency and raise throughput. Perplexity on X ↗
- Funding AI infrastructure provider Nscale is seeking $3.5 billion in pre-IPO financing as it prepares for a potential public listing. TechCrunch ↗
- AIGC Character.AI added preset wallpapers that creators can assign to characters, making the selected background visible to everyone chatting with them. Character.AI on X ↗
- Research Anthropic said Claude agents produced the first complete Lean-verified formalization of Fermat’s Last Theorem in 11 days, spanning 13 million lines. Anthropic ↗
- Agents Google expanded Gemini’s background-running Daily Brief feature to more free Gemini app users in the United States. Gemini App on X ↗
- Agents AWS published guidance for configuring expiration and deletion policies for AgentCore memory, including strategies based on sessions, users and retention requirements. AWS Machine Learning Blog ↗
- AIGC Runway introduced a self-service Team plan supporting two to nine seats, pooled credits, shared projects and centralized workspace administration. Runway on X ↗
- Open Source Qdrant released version 1.19.1, claiming quantized HNSW searches are approximately 1.5 to 2.5 times faster alongside parallelization improvements. Qdrant on X ↗
Microsoft Adds Fast MAI-Image Model to Foundry
MAI-Image-2.6-Flash pairs lower prices with production-focused editing, while the full model leads a major third-party benchmark.
- ⚡ Quick hits
- Research AWS outlined a SageMaker HyperPod factory for building physical-AI models with Nvidia Cosmos 3, covering data preparation, training and evaluation workflows. AWS Machine Learning Blog ↗
- Funding Resect AI emerged with $25 million to develop an enterprise accountability layer that detects, audits and redirects model hallucinations during runtime. SiliconANGLE ↗
- AIGC ElevenLabs partnered with Brazilian ALS nonprofit ABrELA to expand voice-restoration support through its Impact Program for people experiencing voice loss. ElevenLabs on X ↗
GitHub HydraFusion Routes Coding Work Across Models
GitHub’s experimental coding agent selects single, cascade or critique workflows to trade model cost against verified task quality.
Google Expands Lyria 3.5 Across Gemini and API
Google has moved its latest music model into global consumer and developer channels, widening access beyond its earlier Flow Music launch.
- ⚡ Quick hits
- Agents Experian partnered with ServiceNow to connect AI agents with its Ascend data, risk, identity, decisioning and governance capabilities inside enterprise workflows. SiliconANGLE ↗
- Agents Gemini Spark can now edit images, curate albums and execute workflows inside Google Photos for eligible US Pro and Ultra subscribers. TechCrunch ↗
DeepSeek Plans 160,000-Chip Huawei AI Cluster
DeepSeek reportedly plans a vast Ascend deployment in Inner Mongolia, testing whether Chinese hardware can sustain frontier-scale inference.