Vibe Coding
Pattern: Local Models Became a First-Class Feature of Hosted Coding Agents in a Single 48-Hour Window
Three independent moves landed 2026-08-10 through 08-11: GitHub shipped Ollama as a BYOK provider in Copilot for JetBrains, Unsloth Desktop shipped `unsloth start claude` to point Claude Code and Codex at locally-served models, and Meta released Muse Glimmer 30B under Apache 2.0 explicitly benchmarked on MCP-Atlas and SWE-Bench. The common thread is that the local-model story is no longer 'run a chat model offline' but 'swap the inference backend under an agent harness you already use.' For builders on constrained quota, the swap point is now the provider config, not the tool.
↳ Follow the thread