The AI Toolchain — issue -227, January 2, 2026

GitHub Copilot CLI v0.0.374 adds auto-compaction, a /compact command, built-in subagents, and a web_fetch tool.

$ git clone --branch v0.0.374 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v0.0.374

$ /compact

• ›Adds /compact command to manually compact the conversation context.
• ›Adds web_fetch built-in tool for fetching web content directly within a session.
• ›Introduces auto-compaction that triggers automatically at 95% of the token limit to keep sessions alive.
• ›Introduces built-in subagents for exploring and managing tasks.

AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI

Pi v0.31.0 adds session trees with in-place branching, a restructured hooks/tools API, structured compaction with file tracking, and new UI commands.

$ git clone --branch v0.31.0 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:
$ git checkout v0.31.0

import { discoverAuthStorage, discoverModels } from "@mariozechner/pi-coding-agent";

const authStorage = discoverAuthStorage();  // ~/.pi/agent/auth.json
const modelRegistry = discoverModels(authStorage);  // + ~/.pi/agent/models.json

const model = modelRegistry.find("anthropic", "claude-sonnet-4-20250514");
const apiKey = await modelRegistry.getApiKey(model);
const available = await modelRegistry.getAvailable();

pi.registerCommand("branch-here", {
  description: "Branch session from a specific entry",
  async handler(ctx) {
    await ctx.waitForIdle();
    const entries = ctx.sessionManager.getBranch();
    const target = entries[entries.length - 2];  // second-to-last entry
    await ctx.branch(target.id);
  }
});

• ›Sessions now use a tree structure with id/parentId fields, enabling in-place branching via the new /tree command; existing sessions are auto-migrated from v1 to v2 on first load.
• ›New /tree interactive navigator supports search, ←/→ paging, filter modes toggled with Ctrl+O (default → no-tools → user-only → labeled-only → all), and l to bookmark entries as LabelEntry.
• ›New /share command uploads a session as a secret GitHub gist and returns a shareable URL via shittycodingagent.ai.
• ›Adds enabledModels key to settings.json as an allowlist for models, mirroring the --models CLI flag.
• ›New ModelRegistry class (discoverModels, discoverAuthStorage) resolves model discovery and API key lookup from ~/.pi/agent/auth.json and ~/.pi/agent/models.json; exposed to hooks and tools via ctx.modelRegistry.

Qwen Code v0.6.0-nightly adds system-locale-based LLM language auto-detection and ChatCompletionContentPart merging support.

$ git clone --branch v0.6.0-nightly.20260102.105ad743 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.6.0-nightly.20260102.105ad743

• ›Supports merging ChatCompletionContentPart objects and filters empty messages in API content handling.
• ›Auto-detects the preferred LLM output language from the system locale, removing the need to manually configure language preference.
• ›Adds Russian translation strings for internationalized UI surfaces.

KoboldCpp v1.105.4 adds --gendefaults for API payload overrides, a new Adaptive-P sampler, and expanded image/video model support.

$ git clone --branch v1.105.4 https://github.com/LostRuins/koboldcpp.git
# already have the repo? check out this version:
$ git checkout v1.105.4

$ koboldcpp --model mymodel.gguf --gendefaults '{"temperature": 0.7, "top_k": 40}'

$ koboldcpp --model mymodel.gguf --sdmodel wan2.2.gguf --sdvaeauto

• ›Adds --gendefaults flag, accepting a JSON dictionary to append or overwrite any API fields (e.g. temperature, top_k, step count) on all incoming API payloads — useful for correcting misbehaving frontends; supersedes and absorbs the now-deprecated --sdgendefaults.
• ›Adds support for the 'Adaptive-P' sampler, which allows selecting lower-probability tokens; configure via adaptive target and adaptive decay parameters, recommended for use alongside min-P.
• ›Adds support for Ovis Image and Qwen Image Edit model types.
• ›Adds TAEHV support for WAN VAE (compatible with Wan2.2 videos, Qwen Image, and Qwen Image Edit), enabling large memory savings; activate via the 'TAE SD' checkbox or --sdvaeauto.
• ›Adds support for using remote HTTP MCP servers for tool calling in Kobold Lite.

llama.cpp b7614 adds support for the Maincoder-1B model architecture.

$ git clone --branch b7614 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b7614

• ›Adds support for the Maincoder-1B model architecture via a new maincoder.cpp backend.