Kevin Mack’s Post

Get GitHub Copilot directly in your terminal. It takes just seconds to get Copilot CLI running on your machine: 📦 Install via npm, Homebrew, or WinGet 🔐 Authenticate your GitHub account 🚀 Start coding No IDE required. No plugins. Just your terminal and Copilot — ready to code, debug, and explore your codebase from the command line. Credit: @GitHub #GitHub #Copilot #CopilotCLI #AI #DeveloperTools #Terminal #Productivity

Is anyone else noticing how bad github copilot has been over the last 24 hours. First they pull Opus 4.6 and 4.5 for pro and pro+ users last night while i was in the middle of using them. Today Sonnet 4.6 seems to have regressed 2 years. Not in every chat though. Things are going a bit strange.

It seems that Claude models are much cheaper to use with GitHub Copilot than with Claude Code. Has anyone else had the same experience?

ollama launch copilot Ollama now supports GitHub's Copilot CLI, the terminal agent that works directly with repositories on GitHub. You can use it to: Explore issues and PRs. Search across repos by label (e.g. good first issue, help wanted) and bring that context into your session. Plan and scaffold work from a ticket. Hand Copilot CLI an issue and have it map out the change, edit the files, and run the commands to get it done. Navigate unfamiliar codebases. Point it at a repo and ask it to explain the structure, install dependencies, and walk you through how the pieces fit together. Learn more: https://lnkd.in/gYRtvATS Thank you to the Microsoft GitHub team members for building this!

A meaningful update in AI-powered developer workflows: Ollama is now being integrated with the GitHub Copilot CLI from GitHub. This allows developers to run and use local language models directly within CLI-based workflows—bringing more control to how AI is used in development. What this enables: • Local model execution (better control over data and privacy) • Less dependence on cloud-only inference • Flexibility to experiment with different models • Seamless integration into existing CLI workflows This reflects a broader shift toward hybrid AI tooling—combining local and cloud capabilities for more efficient and adaptable development environments. A practical step forward for developers building AI-driven systems. #AI #DeveloperTools #GitHub #Copilot #Ollama #SoftwareEngineering

The ability to point your terminal at a repo to explain structures, or have it scaffold and execute work directly from a ticket, is going to be really usefull.

Never used Ollama in day to day workflow, Now a days LLMs, Agents are the most important part of devs  Every where we are using AI, whether it is making documentation, code, PPTs, discussing about new project, anlyzing things and getting feedback  Claude is now a days too much famous, every where Claude but if we see other models they are also performing well good enough.  If we divide the task I think Ollama also play good I hope so What's you're views on it

This just arrived while I’m installing Ollama to try Gemma 4 and thinking if it works with GitHub Copilot. If it works, it would be a great news. **Update** Integrating Ollama models with GitHub Copilot CLI and GitHub Copilot in VS Code is straightforward, no issues at all. But performance isn't practical unless you invested in hardware. Gemma 4 performed better than Qwen 3.6, because the latter has a support issue with CUDA. Also, it's compact, so its memory requirements are reasonable. Gemma 4 (Size: 9.6 GB) - supports CUDA Input tokens eval rate: 50.67 TPS Output tokens eval rate: 18.87 TPS Qwen 3.6 (Size: 23 GB) - no CUDA support Input tokens eval rate: 9.34 TPS Output tokens eval rate: 8.00 TPS #GitHubCopilot #Ollama

This is the direction enterprise AI adoption should be heading. Running Copilot against local models via Ollama solves one of the biggest blockers for regulated industries and data-sensitive environments: your code never leaves your infrastructure. For organizations in real estate, finance, or government — where IP and client data can't touch external inference endpoints — this changes the calculus entirely. We've been cautious about cloud-based coding assistants for exactly this reason. Local inference with the same developer experience as cloud tools isn't a compromise anymore — it's a viable enterprise path. The question IT leaders need to be asking now: do you have the GPU infrastructure and MLOps capability to actually run this at scale? Because the tooling is ready. The gap is internal readiness.

I wrote about a capability in GitHub Copilot CLI that I think is especially powerfull: it can orchestrate multiple models and subagents in parallel, then consolidate the results into one review. The core idea is simple: even strong models have different blind spots. Running more than one model can surface issues that a single pass might miss, and Copilot CLI makes that workflow practical. 👇