The AI Toolchain — issue -202, January 27, 2026

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

Claude Code v2.1.20 adds PR review status in the footer, vim history nav, TaskUpdate deletions, and per-directory CLAUDE.md loading.

$ git clone --branch v2.1.20 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:
$ git checkout v2.1.20

$ CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD=1 claude --add-dir /path/to/shared-config

• ›Adds arrow key history navigation in vim normal mode when the cursor cannot move further.
• ›Adds external editor shortcut (Ctrl+G) to the help menu for easier discovery.
• ›Adds PR review status indicator to the prompt footer showing the current branch's PR state (approved, changes requested, pending, or draft) as a colored dot with a clickable link.
• ›Supports loading CLAUDE.md files from additional directories specified via --add-dir (requires CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD=1).
• ›Adds ability to delete tasks via the TaskUpdate tool.

Autonomous coding agent as an SDK, IDE extension, or CLI assistant.

Cline v3.54.0 adds native tool calls for Ollama and switches Amazon Bedrock default to Sonnet 4.5.

$ git clone --branch v3.54.0 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout v3.54.0

• ›Adds native tool calls support for the Ollama provider.
• ›Sonnet 4.5 is now the default Amazon Bedrock model ID.

Glamourous agentic coding for all

Crush v0.36.0 adds crush stats usage dashboard, session deletion, session title editing, and multi-file drag & drop.

$ git clone --branch v0.36.0 https://github.com/charmbracelet/crush.git
# already have the repo? check out this version:
$ git checkout v0.36.0

$ crush stats

• ›New crush stats command displays a per-project usage dashboard with token and session statistics.
• ›Adds ability to delete sessions directly from the UI.
• ›Enables renaming/updating session titles.
• ›Supports drag & drop of multiple files at once, including on Windows.

GitHub Copilot CLI v0.0.396 adds copilot version, copilot update, /init, /experimental, and preToolUse hook argument control.

$ git clone --branch v0.0.396 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v0.0.396

$ copilot update

$ /plugin install github.com/example-org/copilot-plugin

$ /init

• ›Adds copilot version and copilot update commands for version inspection and self-updating from the CLI.
• ›Adds /experimental command and --experimental flag to opt into experimental features.
• ›Adds /init command to generate Copilot instructions for a project.
• ›Adds preToolUse hooks that can now deny tool execution and modify arguments before a tool runs.
• ›Extends /plugin install to support GitHub repos, URLs, and local paths as install sources.

Lightweight coding agent that runs in your terminal

Codex CLI v0.92.0 adds dynamic tool injection, thread unarchive RPC, MCP OAuth scopes config, multi-agent explorer role, and cached web_search by default.

$ git clone --branch rust-v0.92.0 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.92.0

• ›Enables dynamic tool injection at API v2 thread startup, routing tool calls and responses end-to-end through the server and core tool pipeline.
• ›Adds thread/unarchive RPC to restore archived rollouts back into active sessions.
• ›Supports defining OAuth scopes for MCP servers directly in config.toml, eliminating the need to pass --scopes on every login.
• ›Expands multi-agent collaboration with an explorer role, improved collab event mapping, and max-depth guardrails to cap agent recursion.
• ›Adds thread list filtering in the app server to simplify browsing large thread sets.

Qwen Code SDK TypeScript v0.1.4 adds Japanese and Portuguese locales, extensions detail command, plugin marketplace UI, and /bug in non-interactive mode.

$ git clone --branch sdk-typescript-v0.1.4 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout sdk-typescript-v0.1.4

$ qwen /bug

$ qwen extensions detail <extension-name>

• ›Adds /bug command availability to non-interactive mode.
• ›Adds detail subcommand for extensions with improved extension validation via feat(extensions): add detail command.
• ›Adds settings support for experimental skills under the experimental namespace in CLI settings.
• ›Adds ui.showLineNumbers setting respected by the DiffRenderer for diff output.
• ›Adds plugin selection UI for the Claude marketplace within the extensions workflow.

$ git clone --branch v0.8.1 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.8.1

$ git clone --branch v0.8.0 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.8.0

$ git clone --branch v0.8.0-preview.4 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.8.0-preview.4

$ git clone --branch v0.8.0-preview.3 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.8.0-preview.3

$ qwen /bug

$ qwen extensions detail <extension-name>

$ git clone --branch v0.8.0-nightly.20260127.eef9c49e https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.8.0-nightly.20260127.eef9c49e

Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.

CrewAI 1.9.0 adds A2A task execution, Keycloak SSO, structured outputs, and native multimodal file handling.

$ git clone --branch 1.9.0 https://github.com/crewAIInc/crewAI.git
# already have the repo? check out this version:
$ git checkout 1.9.0

from pydantic import BaseModel
from crewai import Agent, Task, Crew

class ThreatReport(BaseModel):
    severity: str
    affected_systems: list[str]
    recommendation: str

analyst = Agent(role="Threat Analyst", goal="Analyze security incidents", backstory="...", verbose=False)
task = Task(
    description="Analyze the attached incident log and produce a threat report.",
    expected_output="A structured threat report",
    agent=analyst,
    response_format=ThreatReport,
)
crew = Crew(agents=[analyst], tasks=[task])
result = crew.kickoff()
report: ThreatReport = result.pydantic

• ›Adds structured outputs and response_format support across providers for typed, schema-constrained LLM responses.
• ›Adds response_id field in streaming responses for tracking and correlating streamed completions.
• ›Adds event ordering and parent-child hierarchy to the event system for richer execution tracing.
• ›Adds Keycloak SSO provider support for enterprise authentication integration.
• ›Adds native multimodal file handling with OpenAI Responses API support.

Haystack v2.23.0 adds human-in-the-loop agent confirmation strategies, image-returning tools, and automatic custom-component serialization.

$ git clone --branch v2.23.0 https://github.com/deepset-ai/haystack.git
# already have the repo? check out this version:
$ git checkout v2.23.0

from haystack.components.agents import Agent
from haystack.components.generators.chat import OpenAIResponsesChatGenerator
from haystack.tools import ComponentTool
from haystack.dataclasses import ChatMessage, ImageContent
from haystack import component

@component
class ImageRetriever:
    @component.output_types(images=list[ImageContent])
    def run(self):
        return {"images": [ImageContent.from_file_path("/data/image.jpg")]}

image_tool = ComponentTool(
    component=ImageRetriever(),
    outputs_to_string={"raw_result": True, "source": "images"}
)

agent = Agent(
    chat_generator=OpenAIResponsesChatGenerator(model="gpt-5-nano"),
    system_prompt="Retrieve images and describe them.",
    tools=[image_tool],
)
result = agent.run(messages=[ChatMessage.from_user("Retrieve the image and describe it.")])
print(result["last_message"].text)

import json

def save_to_db(snapshot: dict) -> None:
    db.snapshots.insert_one({"data": json.dumps(snapshot)})

result = pipeline.run(
    data={"query": "What is RAG?"},
    snapshot_callback=save_to_db
)

• ›Adds confirmation_strategies parameter to Agent, accepting per-tool BlockingConfirmationStrategy instances driven by AlwaysAskPolicy, AskOncePolicy, or NeverAskPolicy, with pluggable UIs (RichConsoleUI, SimpleConsoleUI) — enabling agents to pause for human approval before executing tools.
• ›Expands ToolCallResult.result to accept lists of TextContent and ImageContent objects, allowing tools to return images to providers such as OpenAIResponsesChatGenerator and AnthropicChatGenerator.
• ›Adds raw_result key support to the outputs_to_string parameter of Tool, ComponentTool, and PipelineTool for returning image results without string conversion.
• ›Adds outputs_to_string parameter to create_tool_from_function and the @tool decorator for additional customization of tool output formatting.
• ›Adds snapshot_callback parameter to Pipeline.run() to handle pipeline snapshots with a custom function (e.g., saving to a database or remote service) instead of the default file-saving behavior.

PydanticAI v1.48.0 adds domain allowlisting for WebSearchTool, continuous usage stats for OpenAI, and model_settings support for Mistral streaming.

$ git clone --branch v1.48.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:
$ git checkout v1.48.0

from pydantic_ai.tools.web_search import WebSearchTool

tool = WebSearchTool(allowed_domains=["example.com", "docs.openai.com"])

from pydantic_ai import Agent

agent = Agent(
    "openai:gpt-4o",
    model_settings={"continuous_usage_stats": True},
)

• ›Adds allowed_domains parameter to WebSearchTool to restrict OpenAI web searches to specific domains.
• ›Adds continuous_usage_stats model setting for OpenAI to receive token usage statistics on every streamed chunk.
• ›Applies model_settings to Mistral streaming JSON mode, enabling per-request model configuration during Mistral streaming.

SRE Agent - CNCF Sandbox Project

HolmesGPT 0.19.0 adds GitHub, Confluence, and Splunk MCP servers, image input support, and scheduled prompt investigations.

$ git clone --branch 0.19.0 https://github.com/HolmesGPT/holmesgpt.git
# already have the repo? check out this version:
$ git checkout 0.19.0

• ›Adds Splunk MCP server support, replacing any prior ad-hoc Splunk integration with an explicit MCP-based server.
• ›Adds GitHub MCP server, replacing the legacy GitHub toolset with an MCP-based implementation.
• ›Adds Confluence MCP server for Confluence-based investigations and workflows.
• ›Supports passing an auth token to MCP servers via HTTP header.
• ›Adds support for image input to the chat API, including a new images field in the chat API.

Jan v0.7.6 adds message editing, Anthropic Messages API proxy, search dialog, token usage persistence, and a portable Windows build.

$ git clone --branch v0.7.6 https://github.com/janhq/jan.git
# already have the repo? check out this version:
$ git checkout v0.7.6

• ›Enables proxy to llama.cpp for Anthropic Messages API support, allowing local inference to be addressed via the Anthropic Messages API format.
• ›Persists token usage into message metadata so per-message token counts are retained across sessions.
• ›Adds a search dialog for quickly navigating conversations and content within the app.
• ›Adds a portable Windows build for running Jan without installation.
• ›Adds the ability to edit and delete chat messages directly in the conversation UI.

llama.cpp b7842 enables FP16 MMA kernels on PowerPC CPUs for faster CPU inference.

$ git clone --branch b7842 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b7842

• ›Enables FP16 MMA (matrix multiply-accumulate) kernels for PowerPC CPUs via ggml-cpu, accelerating on-CPU inference on PPC hardware.

Get up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

Ollama v0.15.2 adds ollama launch clawdbot to spin up Clawdbot using local Ollama models.

$ git clone --branch v0.15.2 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.15.2

$ ollama launch clawdbot

• ›New ollama launch clawdbot command launches Clawdbot powered by locally running Ollama models.

Arize Phoenix v12.32.0 adds a tool invocation accuracy metric for evaluating agent tool-calling behavior.

$ git clone --branch arize-phoenix-v12.32.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v12.32.0

• ›Adds a tool invocation accuracy metric to measure how accurately agents invoke tools.

Langfuse v3.149.0 adds an events-based trace/observation table, corrected output rendering, thinking budget for Vertex Gemini, and multi-org access provisioning.

$ git clone --branch v3.149.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.149.0

• ›Adds support for multiple default orgs/projects in automated access provisioning configuration, enabling broader SSO/SCIM onboarding rules.
• ›Adds a thinking budget setting for Vertex Gemini models in the playground.
• ›Adds an events-based observation/trace table with experiment filters, trace ID filter, scores, and session/user ID columns in the UI.
• ›Adds footer rendering for corrected output in the IO Preview panel when corrections are enabled.
• ›Adds preflight checks for delete operations on events, observations, and traces to guard against unsafe deletions.