The AI Toolchain — issue -114, April 27, 2026

Agno v2.6.2 adds a Workspace toolkit giving agents read/write/shell access to a local directory tree with HITL confirmation gates.

$ git clone --branch v2.6.2 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:
$ git checkout v2.6.2

• ›Adds WorkspaceTools toolkit exposing read, list, search, write, edit, move, delete, and shell operations scoped to a root directory tree, with destructive operations gated by Agno's built-in human-in-the-loop confirmation by default.

Build resilient agents.

LangGraph 1.1.10 lets ToolNode tools return mixed lists of Command and ToolMessage objects.

$ git clone --branch 1.1.10 https://github.com/langchain-ai/langgraph.git
# already have the repo? check out this version:
$ git checkout 1.1.10

from langgraph.prebuilt import ToolNode
from langgraph.types import Command
from langchain_core.messages import ToolMessage

def my_tool(tool_call_id: str, query: str) -> list:
    # Emit a state update command AND a tool result message
    return [
        Command(update={"retrieved": query}),
        ToolMessage(content=f"Result for {query}", tool_call_id=tool_call_id),
    ]

node = ToolNode([my_tool])

• ›Enables ToolNode tools to return list[Command | ToolMessage], allowing a single tool call to emit both control-flow commands and tool messages in one response.

Your own personal AI assistant. Any OS.

OpenClaw v2026.4.25 adds TTS overhaul with 6 new providers, expanded OpenTelemetry coverage, browser automation upgrades, PWA/Web Push, and new plugin registry CLI commands.

$ git clone --branch v2026.4.25 https://github.com/openclaw/openclaw.git
# already have the repo? check out this version:
$ git checkout v2026.4.25

$ /tts chat on
# ... chat session ...
/tts chat off

$ openclaw browser start --headless

$ openclaw plugins registry --refresh

• ›Adds /tts latest read-aloud command and /tts chat on|off|default session-scoped auto-TTS controls for WhatsApp and chat channels.
• ›Adds Azure Speech as a bundled TTS provider with Speech-resource auth, SSML escaping, native Ogg/Opus voice-note output, and telephony output.
• ›Enables per-agent TTS voice overrides via agents.list[].tts and per-account deep-merge of channels.<channel>.accounts.<id>.tts for Feishu and QQBot.
• ›Adds channels.discord.voice.model config key to override the LLM used for Discord voice channel responses.
• ›Adds Google Meet calendar-backed attendance export workflows, export manifests, and dry-run previews.

Glamourous agentic coding for all

Crush v0.63.0 introduces PreToolUse hooks to intercept, deny, or rewrite tool calls before execution.

$ git clone --branch v0.63.0 https://github.com/charmbracelet/crush.git
# already have the repo? check out this version:
$ git checkout v0.63.0

• ›Adds PreToolUse hook that runs before any tool call, enabling deny, halt, rewrite of bash commands, or silent side effects.
• ›Introduces built-in crush-hook skill to build, manage, and configure hooks from within Crush.
• ›Supports Claude Code hooks for compatibility with existing hook scripts.

GitHub Copilot CLI v1.0.37 adds shell completion scripts, session sort cycling, and default location-based permission persistence.

$ git clone --branch v1.0.37 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v1.0.37

$ copilot completion zsh > ~/.zsh/completions/_copilot && source ~/.zshrc

• ›Adds copilot completion <bash|zsh|fish> subcommand to generate static shell completion scripts covering subcommands, flags, and known choice values.
• ›Location-based permission persistence is now enabled by default, carrying approvals across sessions for the same directory without re-approval.
• ›Press s in the session picker to cycle sort order across four modes: relevance, last used, created, or name.
• ›/ask responses now render full markdown, including tables and formatted links.
• ›ACP model config options now expose description and metadata fields for clients using the configOptions API.

The open source coding agent.

OpenCode v1.14.27 adds a configurable default shell for terminals and agent shell commands.

$ git clone --branch v1.14.27 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v1.14.27

• ›Adds a configurable default shell for terminals and agent shell commands, controllable via a Desktop setting.

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

Pi v0.70.3 adds self-update via pi update, Azure Cognitive Services endpoint support, and new extension APIs for custom working-state UI.

$ git clone --branch v0.70.3 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:
$ git checkout v0.70.3

$ pi update

• ›Adds self-update capability to pi update so Pi itself can be upgraded alongside installed packages.
• ›Adds Azure Cognitive Services endpoint support for Azure OpenAI Responses deployments.
• ›Adds warnings.anthropicExtraUsage setting in /settings to suppress the Anthropic extra-usage billing warning.
• ›Adds ctx.ui.setWorkingVisible() extension API, enabling extensions to hide the built-in loader row and render a custom working state.

ONNX Runtime 1.25.1 adds LinearAttention, CausalConvState, RotaryEmbedding, and RMSNorm operators with Qwen3.5 WebGPU support.

$ git clone --branch v1.25.1 https://github.com/microsoft/onnxruntime.git
# already have the repo? check out this version:
$ git checkout v1.25.1

• ›Adds LinearAttention and CausalConvState operators to support Qwen3.5 model inference.
• ›Adds RotaryEmbedding (RotEMB) and RMSNorm operators.
• ›Adds Linear Attention signature support.
• ›Updates Reshape to opset 25 and Transpose to opset 24.
• ›Enables Qwen3.5 model execution on the WebGPU execution provider.

llama.cpp server router now forwards form-data, enabling /v1/audio/transcriptions in router mode.

$ git clone --branch b8952 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8952

• ›Enables the server router to forward multipart/form-data requests to model servers, unlocking use of the /v1/audio/transcriptions endpoint in router mode.

vLLM v0.20.0 adds DeepSeek V4, TurboQuant 2-bit KV cache, FA4 default MLA prefill, online quantization frontend, and CUDA 13.0 as default.

$ git clone --branch v0.20.0 https://github.com/vllm-project/vllm.git
# already have the repo? check out this version:
$ git checkout v0.20.0

$ uv pip install vllm --torch-backend=cu129

$ vllm serve Qwen/Qwen3-VL-7B-Instruct --enable-vit-cuda-graph

$ VLLM_MEDIA_CACHE=1 vllm serve <model>

• ›Adds --enable-vit-cuda-graph flag for VLM examples to enable ViT full CUDA graph for video inference.
• ›Adds VLLM_MEDIA_CACHE opt-in environment variable for media URL caching.
• ›Adds request_id field on FinishedRequestStats for per-request completion tracking.
• ›Adds --torch-backend=cu129 recommended install option via uv for CUDA 12.9 users.
• ›Adds donate_graph_module=True support for standalone_compile in torch.compile integration.

Langfuse v3.171.0 allows annotation-sourced scores via the public API and adds GPT-5.5 model pricing.

$ git clone --branch v3.171.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.171.0

$ curl -X POST https://<your-langfuse-host>/api/public/scores \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer <secret-key>' \
  -d '{"traceId": "<trace-id>", "name": "quality", "value": 1, "source": "ANNOTATION"}'

• ›Supports source=ANNOTATION on POST /api/public/scores, enabling human annotation scores to be submitted directly through the public scores API.
• ›Adds GPT-5.5 to built-in model price definitions for automatic cost tracking.