Agent-Scale Git ⚖️, vLLM Accuracy 🎯, PyTorch 2026 🔥

GitHub is replacing repository-wide write serialization with per-reference quorums to overcome cross-region latency during its Azure migration. ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR Together With TLDR DevOps 2026-10-28 Add RAG to your Elastic stack (Sponsor) Your data is already in Elasticsearch — use Elastic Cloud to power Retrieval Augmented Generation (RAG). Skip the new database and add vector storage and LLM-powered search without changing your stack. Reuse your data layer: Store vector embeddings directly in Elasticsearch indices alongside your existing data — no separate vector database required Wire up RAG end to end: Connect Elastic Cloud, LangChain, and Amazon Bedrock with a working Python pipeline from document ingestion to LLM query Access models through a unified API: Use Elasticsearch's Open Inference API to call Amazon Bedrock models for chat completion and text embeddings Follow the step-by-step tutorial and start asking questions of your data today. Build a RAG pipeline on AWS > 📱 News & Trends Gheist: Attempted Scaled Abuse of Stolen GitHub Tokens Stumbles (9 minute read) Gheist is an automated campaign that plants secret-dumping workflows in repositories using stolen GitHub tokens. It has targeted over 850 compromised accounts since September 15. GitHub's recent security controls blocked the attack in public repositories, but private and enterprise environments remain vulnerable. What Took the Internet Offline in Q3 2026 (8 minute read) Cloudflare Radar recorded 49 Internet disruptions during the third quarter, including 19 government-directed shutdowns and 14 power outages. The incidents illustrate dependencies beyond network redundancy. 🚀 Opinions & Tutorials Why Kubernetes autoscalers keep fighting each other (and why the solution must be autonomous) (7 minute read) Nikhil Kurup argues that HPA, VPA, KEDA, and Karpenter mutate the same workload fields from independent loops, producing four failure modes, including a VPA and HPA death spiral and GitOps drift ping-pong. He calls for a unified, predictive autoscaler with explainable decisions. How to Build Golden Paths in Platform Engineering (19 minute read) Spacelift defines a golden path as an opinionated, documented, self-service route for a common task. The guide lists five building blocks and four rollout phases, with a worked Postgres example on AWS built as an rds-postgres module using OpenTofu 1.10.3. Production gets Multi-AZ and deletion protection. The Bridge to Agent-Scale Git Infrastructure (8 minute read) GitHub is replacing repository-wide write serialization with per-reference quorums to overcome cross-region latency during its Azure migration. Internal tests showed 3–4x higher throughput for writes to different references, while an object-storage-backed design allows read capacity to scale independently. 🧑‍💻 Resources & Tools How do you respond when your AI tooling and models fail? (Sponsor) Engineers from incident.io and Skyscanner discuss what happens when AI tooling becomes a single point of failure, and what both teams are doing about it. Watch Now→ Agent-Reach (GitHub Repo) Agent-Reach is an open source CLI tool that lets agents read and search platforms like Twitter, Reddit, YouTube, GitHub, and Bilibili without API fees. It handles login-gated sites, video content, and Chinese platforms to provide structured data for AI agents. OpenBot (GitHub Repo) OpenBot is a local-first desktop workspace for persistent AI teammates that runs Codex, Claude, Grok, and Gemini with dedicated workspaces and task queues. It provides a shared desktop channel, file transfers, and an embedded browser that agents can control. A development preview is available for macOS, Windows, and Linux. 🎁 Miscellaneous Before you expand your GPU pool, measure the outcome (8 minute read) Teams should define each workload's outcome and deadline before expanding a GPU pool. Teams can stagger flexible jobs, load test sustainable serving capacity, and track cost per verified outcome with OpenCost before adding GPUs. Fixing Tool-Call Accuracy in vLLM (7 minute read) vLLM 0.31 can achieve 100% tool-call schema accuracy for Kimi and DeepSeek models through constrained decoding, a technique that guarantees schema adherence during token generation. This fixes a reliability issue where vLLM 0.28 produced invalid tool calls 8% of the time, causing agentic workloads to build up nonsensical context. ⚡ Quick Links The AI Software Factory Summit - SF 30% Off Ticket (Sponsor) Explore advanced AI software use cases live in production showcased by Kelsey Hightower and leaders from NVIDIA, GitHub, and SpaceX this Nov 17–18 in SF. Discount code TLDR_30% Claim Your 30% Off Ticket PyTorch Conference 2026: Google and Meta Open-Source TorchTPU for Native PyTorch on TPUs (4 minute read) Google and Meta open-sourced TorchTPU, a native PyTorch backend for Google's TPUs, at PyTorch Conference North America 2026. Observability patterns for AI agents on Amazon EKS (20 minute read) AWS engineers compare two observability patterns for AI agents on Amazon EKS. Scope Your Personal Access Tokens (6 minute read) Supabase personal access tokens can be restricted to specific organizations, projects, permissions, and expiration dates for CI pipelines and other integrations. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/3b85ceef/10 Track your referrals here. Want to advertise in TLDR? 📰 If your company is interested in reaching an audience of devops professionals and decision makers, you may want to advertise with us. Want to work at TLDR? 💼 Apply here, create your own role or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! TLDR is one of Inc.'s Best Bootstrapped businesses of 2025. If you have any comments or feedback, just respond to this email! Thanks for reading, Kunal Desai & Martin Hauskrecht Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR DevOps isn't for you, please unsubscribe.