{"id":4612,"date":"2026-09-26T18:11:15","date_gmt":"2026-09-26T09:11:15","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/2026\/09\/26\/koboldcpp-v1-122-released\/"},"modified":"2026-09-27T17:53:48","modified_gmt":"2026-09-27T08:53:48","slug":"koboldcpp-v1-122-released","status":"publish","type":"post","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/09\/26\/koboldcpp-v1-122-released\/","title":{"rendered":"KoboldCpp v1.122 Released with Built-in Agentic Framework"},"content":{"rendered":"

<\/p>\n

At a Glance<\/h2>\n
\n\n\n\n\n\n\n\n\n\n
Item<\/th>\nValue<\/th>\n<\/tr>\n<\/thead>\n
Repository<\/td>\nLostRuins\/koboldcpp<\/a><\/td>\n<\/tr>\n
Version<\/td>\nv1.122<\/a><\/td>\n<\/tr>\n
Published<\/td>\n2026-09-26<\/td>\n<\/tr>\n
License<\/td>\nAGPL-3.0<\/td>\n<\/tr>\n
Source type<\/td>\nPrimary source (the publisher itself)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

Values determined by this site’s code at collection time. Dates are JST.<\/em><\/p>\n

<\/p>\n

Overview<\/h2>\n

KoboldCpp v1.122, the latest version of the AGPL-3.0 licensed inference engine for running GGUF models with ease, has been released. KoboldCpp operates as a single-binary application and is a C++-based inference environment that includes the browser-accessible KoboldAI UI.<\/p>\n

The most significant change in this version is the integration of the lightweight agent foundation “KoboldCpp Agent”. Designed with 9 built-in tools and keeping system prompt consumption down to around 2k tokens, it enables the creation of autonomous coding and task execution environments using local LLMs with minimal overhead.<\/p>\n

Breaking Changes and Deprecations<\/h2>\n

This release introduces changes to the build script behavior specifications and streamlines several startup flags.<\/p>\n

\n\n\n\n\n\n\n
Item<\/th>\nOld Specification<\/th>\nNew Specification (v1.122 and later)<\/th>\n<\/tr>\n<\/thead>\n
Portable builds via build script koboldcpp.sh<\/code><\/td>\nGenerates portable builds by default<\/td>\nAutomatically optimizes for the execution environment. Creating binaries for other environments requires setting the environment variable KCPP_PORTABLE=1<\/code><\/td>\n<\/tr>\n
Pipeline parallelism flag<\/td>\nManually specified via dedicated flag<\/td>\nFlag removed (automatically enabled when ubatch<\/code> is smaller than batchsize<\/code>)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

Maintainers distributing packages via koboldcpp.sh<\/code> and users setting up build pipelines must explicitly set the environment variable KCPP_PORTABLE=1<\/code> to prevent generated binaries from being locked to the execution machine’s specific optimizations.<\/p>\n

Key Changes<\/h2>\n

Addition of Built-in Agent Feature “KoboldCpp Agent”<\/h3>\n

A lightweight agent framework acting as an alternative to Opencode, Codex, Claude Code, and others has been bundled directly into KoboldCpp. Featuring 9 built-in tools, it keeps total system prompt consumption including all tool definitions down to approximately 2k tokens.<\/p>\n

It can be enabled from the “Admin” tab in the GUI launcher or by specifying --agent<\/code> in the startup options. It supports connecting to external OpenAI Chat Completions-compatible endpoints, and tool extension via the Model Context Protocol (MCP) is also possible by loading an mcp.json<\/code> file (MCP tools execute on the KoboldCpp server side, while Agent tools execute on the agent client side). Three approval modes\u2014on<\/code>, auto<\/code>, and off<\/code>\u2014are available for confirming tool calls. To run the agent practically, an environment with at least 28k+ context length, 8k+ generated token count, and 12GB+ VRAM is recommended.<\/p>\n

Independent ubatchsize Setting and Pipeline Parallelism Automation<\/h3>\n

Similar to llama.cpp, ubatchsize<\/code> (micro-batch size) can now be specified independently of batchsize<\/code>. Accompanying this, a ubatch<\/code> slider has been added to the GUI, and various slider UIs have been compactly rearranged. In addition, conventional flags for pipeline parallelism have been streamlined, changing to a behavior where they are automatically enabled when the ubatch<\/code> value is set lower than batchsize<\/code>.<\/p>\n

Introduction of Model Swap Control Flag --autoswapthreshold<\/code><\/h3>\n

The --autoswapthreshold<\/code> option has been added to specify the size limit that triggers automatic model swapping. This prevents unnecessary memory swaps from occurring when auxiliary smaller models are loaded and used.<\/p>\n

Expansion of Default Context Size and Improved Context Management<\/h3>\n

The startup default context size has been increased, and the handling of maximum output generation tokens when approaching the maximum context length is now controlled more intelligently. Furthermore, changes have been made to return error objects if an error occurs during streaming generation, and keep-alive whitespace padding has been added for tool streaming and text synchronization requests.<\/p>\n

Fixes for Image and Music Generation and Stable Diffusion Integration<\/h3>\n

Upstream sd.cpp<\/code> synchronization has been incorporated, lowering the SD VAE tiling threshold to 512. Additionally, maximum size limits for image previews have been introduced to properly handle pixel space diffusion models, along with support for audio code generation (two-step processing) when using LLMs for lyrics generation, and improvements to music-related UIs.<\/p>\n

Hardware-Specific Bug Fixes and Optimizations<\/h3>\n

The VRAM usage increase issue occurring in P40 environments has been reverted, and smartcache logging suppression during quiet mode operation, fixes to the tool call parser, and updates to Kobold Lite have been implemented.<\/p>\n

Supported Models and Hardware<\/h2>\n

KoboldCpp v1.122 supports a wide variety of hardware environments and modern model architectures. By merging the latest improvements from upstream llama.cpp<\/code> and sd.cpp<\/code>, inference stability and supported ranges have been further enhanced.<\/p>\n

Supported Models and Quantization Formats<\/h3>\n

This version enables inference for a broad range of LLMs (Large Language Models) provided in the GGUF format. Support for modern model architectures continues to be updated through upstream merges.<\/p>\n