llmman now runs Qwen-Image 2.1 locally straight from GGUF in Rust
danielhanchen · x · 2026-09-30
- The open-source tool llmman ("run any agent on any model") can now run Qwen-Image 2.1 locally, loading UnslothAI's GGUF weights directly on ggml, implemented in Rust.
- A single command generates an image, e.g. llmman run qwen-image-2.1 "A manatee in a sunlit lagoon".
- llmman's core pitch: point agents like Claude Code, Codex or OpenCode at local or hosted models; models are stored as OCI images, so they can be transferred between registries (e.g. Hugging Face → your own) like container images with no intermediate local copy.
More from coding & agent
- Cloudflare launches Workers Issues: built-in error monitoring that feeds production failures straight to your coding agent — ritakozlov · 2026-09-30
- Cloudflare Rebuilds Containers for Agent Sandboxes: 648ms Startup, 6x Faster, FS Snapshots — irvinebroque · 2026-09-30
- Outrank claims 2x median clicks across 2,119 sites; devs debate AI SEO automation vs DIY agents — tibo_maker · 2026-09-30
- Claude Opus 5.5 builds an explorable 5-axis CNC machine in the browser — burhop · 2026-09-30
- Vibe-coded Fly Anywhere app built with Opus 5.5, trailer included, now open source — Meer_AIIT · 2026-09-30
- Agents write 741% more code but ship only 30% more software — human review is the new bottleneck — AI Engineer · 2026-09-30