11,840 TPS on a single GPU
Learn how we optimized NVIDIA Nemotron 3.5 Super VL inference with five vLLM configuration changes, nearly doubling throughput from 6,058 to 11,840 tokens/sec on a single GB300 GPU.
Latest
Building Faster AI Agents with Jev Decision Models and Cline
Learn how Jev Decision Models and Cline help build faster, cheaper AI agents with real-time decision-making, browser automation, and automated guardrails.
Cline Agents Harness Unlocks Benchmark-Leading Performance from NVIDIA Nemotron 3.5 Super VL
Explore how NVIDIA’s latest Nemotron 3.5 Super model performs with Cline’s agent harness, delivering benchmark-leading results and powering real-world AI coding workflows.
Cline Desktop: An open-source app for open-weight models
Meet Cline Desktop, an open-source app for open-weight models. Run parallel AI agents, automate recurring tasks, choose from 300+ models, and extend your workflow with plugins, MCP servers, and skills.
How We Migrated 11 Million Users to Cline's Biggest Harness Upgrade
How Cline migrated its VS Code extension for 11 million developers to the new Cline SDK, built a safe A/B rollout, and reduced agent failures by 10x.
DeepSeek wins IMO Gold on 12 cents
Benchmarking eight AI models on IMO 2026 problems in Cline, we found DeepSeek V4 Flash scored 30/42, clearing the gold medal cutoff for just $0.12. See how open-weight models compare with GPT-5.6 Sol, Claude Fable 5, Kimi K3, and more.
Building a code review agent on the Cline loop
Build a code review agent with the Cline SDK using custom tools, guardrails, hooks, and a two-pass review workflow. Learn how to inspect PR diffs, filter false positives, audit tool usage, and post high-signal review comments to GitHub.
Open-sourcing evals for open-weight agents
Learn how Cline evaluates and improves open weight coding agents using Terminal Bench, with practical heuristics for model performance, token efficiency, reasoning, providers, and eval optimization.
NVIDIA Nemotron 3.5 Lightning just landed in Cline
NVIDIA Nemotron 3.5 Lightning is now available in Cline, bringing a 30B MoE model with 3B active parameters, up to 1M context, and high throughput for agentic coding workflows.
Top 5 Open Weight Models That Matter to Developers in 2026
Explore the best open weight models for coding in 2026. Compare Kimi K3, GLM 5.2, DeepSeek V4 Pro and Flash, and MiniMax M3 across benchmarks, pricing, model size, and developer use cases.
Recursive Self Improvement for Coding Agents
See how recursive self-improvement helped Cline optimize Kimi K3 and achieve an 88.8% SOTA score on Terminal-Bench 2.1 at a fraction of the cost.
How to Save Millions by Self-Hosting LLMs
A practical guide to the economics of self-hosting open-weight LLMs. Using Kimi K2.6 and real production traffic from Cline, this post breaks down GPU memory, inference, batching, pricing, and the point at which self-hosting can save millions.
Announcements
ClinePass, Best of Value for Open-Weight models
ClinePass is a low-cost monthly subscription that pairs Cline's agent harness with a curated set of open-weight models and 2-5x the standard API rate limits, across every surface Cline runs on.
Guides
Extending the Cline loop with plugins and hooks
Learn how to extend the Cline agent loop with plugins and hooks to add custom behavior, lifecycle logging, and enforceable guardrails without rebuilding the harness.
Announcements
Introducing Cline SDK: the upgraded agent runtime, and we rebuilt Cline upon it
0:00 /0:41 1× Before “agents” became a buzzword, Cline was the first real agentic coding experience. Cline started with the VSCode extension and helped a generation of developers step into AI coding. It was a great VS Code extension, but as the technology evolves, it also taught us
Guides
Cline mobile: how to vibe code from your phone
An engineer at Cline told me he was vibe coding on a road trip from his phone. His Mac was at home running Cline agents; he was checking on them from the passenger seat. Here's the setup. What you need Your phone and your Mac both need Tailscale
Guides
Three AIs enter. One survives. What a SIGKILL race reveals about inference speed
We built an arena where three AI coding agents fight to the death. Each agent runs on different hardware, a different inference stack, and a different economic model. They all receive the same task: write a bash script that kills your opponents, then execute it immediately. The last process standing
Insights
Architects or Tenants: Modern AI Stacks Are Being Built on Rented Foundations
You're building on infrastructure you don't control, can't audit, and can't see degrading in real time. The Invisible Dependency For engineers, the inference vendor problem starts with a deceptively simple question: what happens when the model changes and you don't
Prompt Fundamentals
20 one-shot prompts that turn Kanban into an autonomous coding machine
We put together 20 starter prompts for the Kanban sidebar chat. They create linked dependency chains, maximize parallel agent execution, and produce real, working code. Install with npm i -g cline.
Announcements
Announcing Cline Kanban: a CLI-agnostic app for multi-agent orchestration.
Here’s the elephant in the room about coding in 2026: the bottleneck isn’t the AI; it’s you. Not your skill. Not your prompts. Your attention. Your cognitive bandwidth. If you’ve spent any real time with coding agents, you know the feeling. You start the morning with
The Invisible Trap: The decisions quietly transferring control of your stack to model providers
Why cheaper compute won't mean cheaper AI, and what your stack is risking right now. OpenAI projected operating losses of $74 billion in 2028 alone, before an expected pivot to profitability around 2030. Deutsche Bank research calculates that OpenAI's projected cumulative cash burn could exceed $200
Prompts as playbooks; how infrastructure teams codify operational knowledge
Every infrastructure engineer I know runs prompts in their head. They just don't call them that. You're deploying a new service to production. Without thinking about it, you follow a known sequence to ensure a service will run without impacting existing production workloads. You've
One API key for Claude, Gemini, GPT, and everything else
If you've built anything on top of AI models in the last year, you know the routine. Sign up for Anthropic's API, generate a key, store it somewhere safe. Do the same for OpenAI. Again for Google. Each provider has its own billing dashboard, its own
Announcements
A practical guide to hill climbing
We didn't have benchmark numbers, so over a weekend we ran Cline against 89 coding tasks, diagnosed every failure, and shipped fixes that took our score from 47% to 57%. Here's the hill climbing process so you can do it too.