Issues

• [Feature] Add mcp support for the Deep Research feature

  feature request
  unslothai/unsloth#12537 · Filter by author Gitshop77 opened on Oct 2, 2026

• [Feature] Add Code tool for the vision capable LLM to automatically import an image.

  feature request
  unslothai/unsloth#12534 · Filter by author MODAT7g opened on Oct 2, 2026

• [Bug] Unsloth Studio minor bug with the upper right hand corner Context Length meter.

  bugfeature request
  unslothai/unsloth#12533 · Filter by author MODAT7g opened on Oct 2, 2026

• [Feature] Allow Customisation of Sandbox Virtual Memory Limit

  feature request
  unslothai/unsloth#12520 · Filter by author largeapple52 opened on Oct 2, 2026

• [Bug] Constantly Getting "Generation stopped making progress"

  bugfeature request
  unslothai/unsloth#12518 · Filter by author largeapple52 opened on Oct 2, 2026
  2

• Sandbox nul file breaks tools

  unslothai/unsloth#12473 · Filter by author hoogsteders-netizen opened on Oct 1, 2026

• [Feature] Studio Images: select text encoder / engine when downloading Qwen-Image-2.1 (avoid forced ~17GB dense TE)

  unslothai/unsloth#12470 · Filter by author EricPlayZ opened on Oct 1, 2026
  1

• [Bug] Tensor split decode up to 2.9x slower since b10715-mix-86bd2d3 possibly "max_cuda_graphs = 64" from #144 (2x RTX 5070 Ti, Windows/WSL2/Linux)

  unslothai/unsloth#12468 · Filter by author floewe opened on Oct 1, 2026
  3

• [Bug] Companion-device mask widening (#11823) is skipped when the model has saved gpu_ids, so --mmproj-device CUDA1 is still rejected

  bugfeature request
  unslothai/unsloth#12467 · Filter by author ChexterWang opened on Oct 1, 2026

• Unsloth Studio backend: Xet health probe stubs out Triton process-wide, breaking diffusers/xformers ("'function' object has no attribute 'fn'")

  unslothai/unsloth#12466 · Filter by author steza73 opened on Oct 1, 2026

• [Unsloth Bug] Studio: Qwen-Image-2.1 GGUF generation fails - 1-D norm weights never dequantized in no-conversion load path (4096 vs 8192)

  bugfeature request
  unslothai/unsloth#12445 · Filter by author WolCarlos opened on Oct 1, 2026
  2

  

• [Bug] Desktop: Hugging Face quant discovery blocks loading On Device models offline

  unslothai/unsloth#12415 · Filter by author ccaprani opened on Oct 1, 2026
  2

  

• [Unsloth Studio] UNSLOTH_LLAMA_FORCE_COMPILE=1 source build gets an unusable RUNPATH -- fails immediately with "cannot open shared object file"

  unslothai/unsloth#12392 · Filter by author italopaesano opened on Oct 1, 2026

• unsloth_zoo's torch._grouped_mm support probe segfaults the whole process on ROCm (not a catchable exception)

  unslothai/unsloth#12391 · Filter by author italopaesano opened on Oct 1, 2026

• [Unsloth Bug] Studio pages mmproj-F16.gguf from disk during generation — severe t/s regression since latest update; extra args shadow-stripped and --mlock rejected

  unslothai/unsloth#12372 · Filter by author vingaut opened on Sep 30, 2026
  2

• [Feature Request] Auto-chunking or background processing for large text attachments

  feature request
  unslothai/unsloth#12369 · Filter by author mohammedalsayegh opened on Sep 30, 2026

• Add capability to swap between multiple System Prompts

  feature request
  unslothai/unsloth#12366 · Filter by author Aztexx opened on Sep 30, 2026

• When using multiple accounts to access the same local model it - current loaded model does not sync correctly

  bugfeature request
  unslothai/unsloth#12365 · Filter by author Aztexx opened on Sep 30, 2026
  2

• [Studio] OpenAI-compatible API adds a fixed ~1.2 s per request — 3-5x slower than the bundled llama-server on short-text workloads

  unslothai/unsloth#12364 · Filter by author cnlove7777777-art opened on Sep 30, 2026
  

• [Feature] Auto read-aloud after response completes

  feature request
  unslothai/unsloth#12303 · Filter by author deekzzz opened on Sep 29, 2026

• [Feature] Make Unsloth Studio Web UI installable as a PWA

  feature request
  unslothai/unsloth#12262 · Filter by author AshishMBakshi opened on Sep 29, 2026

• [Bug] Unsloth Loading a repo id that exists in a registered scan folder downloads it into the hub cache instead of using the local copy

  bugfeature request
  unslothai/unsloth#12253 · Filter by author VTplus opened on Sep 28, 2026

• [Unsloth Studio] Prebuilt ROCm llama.cpp binary crashes with "out of memory" (hipStreamCreateWithFlags); source build of the same commit works fine

  unslothai/unsloth#12205 · Filter by author italopaesano opened on Sep 28, 2026

• [Feature] Memory across conversations with different models

  feature request
  unslothai/unsloth#12183 · Filter by author Rafael-Silva-Oliveira opened on Sep 28, 2026

• [Feature] Support Gigatoken as a tokenizer backend

  feature request
  unslothai/unsloth#12072 · Filter by author ThongAccount opened on Sep 27, 2026