Ideas for building
better software.

🐯 Liger GRPO meets TRL

Vercel security roundup: improved bot defenses, DoS mitigations, and insights

Since February, Vercel blocked 148 billion malicious requests from 108 million unique IP addresses. Every deployment automatically inherits these protections, keeping your workloads and enabling your team to focus on shipping rather than incidents. Our real-time DDoS filtering, managed Web Application Firewall (WAF), and enhanced visibility ensure consistent, proactive security.secure by default Here's what's new since February. Read more

Middleware insights now available in Vercel Observability

The dashboard now includes a dedicated view for middleware, showing invocation counts and performance metrics.Vercel Observability Observability Plus users get additional insights and tooling: or learn more about and .View the dashboardObservabilityObservability Plus Read more Analyze invocations by request path, matched against your middleware config Break down middleware actions by type (e.g., redirect, rewrite) View rewrite targets and frequency Query middleware invocations using the q

Rate limiting now available on Hobby, with higher included usage on Pro

Rate limiting now has higher included usage and broader availability to help protect your applications from abuse and manage traffic effectively. The first 1,000,000 allowed rate limit requests per month are now included. Hobby teams also get 1 free rate limit rule per project, up to the same included allotment. These changes are now effective and have been automatically applied to your account. Learn more about or now.configuring rate limits create a new rate limiting rule Read more

Faster CDN proxying to external origins

We’ve optimized connection pooling in our CDN to reduce latency when connecting to external backends, regardless of traffic volume. This is now live across all Vercel deployments at no additional cost. Read more : Improved connection reuse and TLS session resumption reduce response times by up to 60% in some regions, with a 15–30% average improvement.Lower latency : 97% connection reuse and more efficient TLS resumption significantly cut the number of new handshakes required.Reduced origin loa

Analytics - New GraphQL Analytics API Explorer and MCP Server

We’ve launched two powerful new tools to make the GraphQL Analytics API more accessible: GraphQL API Explorer The new GraphQL API Explorer helps you build, test, and run queries directly in your browser. Features include: In-browser schema documentation to browse available datasets and fields Interactive query editor with autocomplete and inline documentation A "Run in GraphQL API Explorer" button to execute example queries from our docs Seamless OAuth authentication — no manual setup required

Addendum to OpenAI o3 and o4-mini system card: OpenAI o3 Operator

We are replacing the existing GPT-4o-based model for Operator with a version based on OpenAI o3. The API version will remain based on 4o.

Tiny Agents in Python: a MCP-powered agent in ~70 lines of code