Issues

• Router Mode + RPC: Model swap times out after 120 seconds when loading a large model over RPC

  LostRuins/koboldcpp#2533 · Filter by author STRIEH opened on Oct 12, 2026
  1

• Tool call request disappears on page reload

  LostRuins/koboldcpp#2528 · Filter by author NGgmnBQGwR opened on Oct 11, 2026
  1

• Token-rule grammars (<[id]>, !<[id]>) return empty output or abort with GGML_ASSERT(!grammar->stacks.empty())

  LostRuins/koboldcpp#2525 · Filter by author reithan-local-agent opened on Oct 10, 2026
  1

• RX 580 not working via vulkan+windows

  LostRuins/koboldcpp#2522 · Filter by author widardd opened on Oct 8, 2026
  11

• [Feature Request] Save/load KV state slots to a file, so long chats stay warm on low-RAM machines with fast NVMe drives

  LostRuins/koboldcpp#2520 · Filter by author linux4life1 opened on Oct 5, 2026

• LibreChat native web_search tool causes KeyError: 'name' in compress_tools_array

  LostRuins/koboldcpp#2516 · Filter by author neonmaus opened on Oct 3, 2026
  2

• [Feature Request]: Reset agent turns when confirming an action

  LostRuins/koboldcpp#2515 · Filter by author neph1 opened on Oct 3, 2026
  1

• [Feature Request] When clicking "Load Model" in the GUI launcher, do not close the GUI launcher, and add a "Unload Model" button.

  LostRuins/koboldcpp#2513 · Filter by author vgame86 opened on Sep 30, 2026
  1

• A tool with no properties is rendered so the model must invent an empty parameter name (--jinja_tools, 1.121)

  LostRuins/koboldcpp#2510 · Filter by author firedesiresa-bit opened on Sep 29, 2026
  2

• Using cli with .kcpps that has cli set to false results in "Overriding Config Value: cli" and EOFError

  LostRuins/koboldcpp#2506 · Filter by author jojorne opened on Sep 28, 2026
  6

• Tool calling: Just Yes/No with no way to escape.

  LostRuins/koboldcpp#2505 · Filter by author rustysixslinger opened on Sep 28, 2026
  2

• [ Feature Request ] Precompiled binary for Nvidia GB10

  LostRuins/koboldcpp#2503 · Filter by author dvin963 opened on Sep 27, 2026
  2

• 100% cpu spinning on tool call, generastion cannot be aborted

  LostRuins/koboldcpp#2500 · Filter by author schmorp opened on Sep 26, 2026
  4

• [ Feature Request ] Update for Qwen image 2.1 support.

  LostRuins/koboldcpp#2495 · Filter by author Peterx2026 opened on Sep 24, 2026
  5

• RX7900 GRE / Windows 11 / Gemma4 Model works in 1.119 (nocuda) but not in 1.121 (nocuda)

  LostRuins/koboldcpp#2494 · Filter by author 8u6man opened on Sep 22, 2026
  2

• Prompt processing hangs indefinitely when prompt token count lands a few tokens past the batch/ubatch size (reproduced at batchsize=1024 → hang at ~1030 tokens, and batchsize=1536 → hang at ~1542 tokens)

  LostRuins/koboldcpp#2493 · Filter by author arnatoatadaink opened on Sep 22, 2026
  2

• --autofitpadding for Secondary GPUs

  LostRuins/koboldcpp#2492 · Filter by author jobobby04 opened on Sep 21, 2026
  2

• [Bug] --routermode + --config triggers unnecessary model reload on first API request

  LostRuins/koboldcpp#2490 · Filter by author VL4DST3R opened on Sep 21, 2026
  1

• [Bug]: Visual artifacts appear on application window upon launch

  LostRuins/koboldcpp#2489 · Filter by author FckingChuck opened on Sep 21, 2026
  3

• [Feature Request] Auto-save config to kcpps when launching

  LostRuins/koboldcpp#2482 · Filter by author Sydebrix opened on Sep 19, 2026
  4

• [FIXED][Bug] Makefile LDFLAGS for LLAMA_VULKAN is shadowed by misplaced endif

  LostRuins/koboldcpp#2476 · Filter by author tncfdp opened on Sep 17, 2026
  6

• [Feature Request] Please add a UI toggle for DP2A emulation (DISABLE_DP4A) to fix slow generation on NVIDIA CMP cards

  LostRuins/koboldcpp#2472 · Filter by author zotyfa-web opened on Sep 15, 2026
  4

• [BUG?] - MMProj and MTP don't work if Autofit for qwen4exp?

  LostRuins/koboldcpp#2471 · Filter by author SabinStargem opened on Sep 15, 2026
  6

• More n-gram control (For Qwen 3.8 next and following)

  LostRuins/koboldcpp#2470 · Filter by author 435jonas opened on Sep 15, 2026
  1

• DeepSeek V4 Flash is classified as "RNN or Hybrid" and forced into SmartCache lifeboat mode, but plain fast-forward works fine with --noshift

  LostRuins/koboldcpp#2464 · Filter by author Rw5xJ7y opened on Sep 13, 2026
  2