The AI Toolchain — issue -149, March 23, 2026

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

CrewAI 1.11.1 adds flow_structure() serializer for programmatic Flow class introspection.

$ git clone --branch 1.11.1 https://github.com/crewAIInc/crewAI.git
# already have the repo? check out this version:
$ git checkout 1.11.1

flow = MyFlow()
structure = flow.flow_structure()
print(structure)

• ›Adds flow_structure() serializer method to the Flow class for structured introspection of flow topology.

OpenAI Agents SDK v0.13.0 adds MCP resource methods, streamable HTTP session resumption, and opt-in reasoning-content replay for Chat Completions.

$ git clone --branch v0.13.0 https://github.com/openai/openai-agents-python.git
# already have the repo? check out this version:
$ git checkout v0.13.0

session_id = mcp_server.session_id

model = LiteLLMModel(..., should_replay_reasoning_content=True)

resources = await mcp_server.list_resources()
content = await mcp_server.read_resource(resources[0].uri)

• ›Adds list_resources(), list_resource_templates(), and read_resource() methods to MCPServer, exposing MCP resource access directly from the SDK.
• ›Adds session_id property to MCPServerStreamableHttp, enabling streamable HTTP sessions to be resumed across reconnects or stateless workers.
• ›Adds should_replay_reasoning_content opt-in flag to Chat Completions integrations, improving reasoning/tool-call continuity for adapters such as LiteLLM and DeepSeek.
• ›Changes the default Realtime WebSocket model to gpt-realtime-1.5, so new Realtime agent setups use the newer model without extra configuration.

Your own personal AI assistant. Any OS.

OpenClaw v2026.3.23 adds pay-as-you-go Qwen endpoints, Knot theme overhaul, SHA-256 CSP hashing, and cron timezone-aware scheduling.

$ git clone --branch v2026.3.23 https://github.com/openclaw/openclaw.git
# already have the repo? check out this version:
$ git checkout v2026.3.23

$ openclaw cron add --at '2026-11-01T09:00' --tz America/New_York 'openclaw run my-morning-task'

• ›Adds standard pay-as-you-go DashScope endpoints for China and global Qwen API keys under the relabeled Qwen (Alibaba Cloud Model Studio) provider group.
• ›New Knot UI theme with black-and-red palette meeting WCAG 2.1 AA contrast, discrete roundness stops, config icons for Diagnostics/CLI/Secrets/ACP/MCP sections, and improved aria-labels.
• ›Adds SHA-256 hash computation for inline <script> blocks in index.html, included in the script-src CSP directive to allow only explicitly hashed bootstrap code.
• ›Adds --tz <iana> support to openclaw cron add|edit --at for correct local wall-clock scheduling of one-shot datetimes, including DST boundaries.
• ›Adds asDocument as a user-facing alias for forceDocument on Telegram image and GIF sends.

$ git clone --branch v2026.3.22 https://github.com/openclaw/openclaw.git
# already have the repo? check out this version:
$ git checkout v2026.3.22

$ openclaw plugins install clawhub:exa

plugins:
  entries:
    exa:
      config:
        webSearch:
          extractContent: true
          searchMode: neural

agents:
  defaults:
    imageGenerationModel:
      primary: "google/gemini-3-pro-image-preview"

browser-use 0.12.3 ships CLI 2.0 with direct CDP, ~50ms latency, LiteLLM support, and multi-session browser control for AI coding agents.

$ git clone --branch 0.12.3 https://github.com/browser-use/browser-use.git
# already have the repo? check out this version:
$ git checkout 0.12.3

$ browser-use --profile "Default" open https://github.com
browser-use state
browser-use click 2
browser-use screenshot page.png

$ browser-use -s work open https://work.example.com
browser-use -s personal open https://gmail.com
browser-use sessions
browser-use close --all

$ browser-use tunnel 3000
# outputs: https://abc.trycloudflare.com
browser-use open https://abc.trycloudflare.com
browser-use state
browser-use screenshot --full result.png
browser-use tunnel stop --all

• ›Adds browser-use --connect open <url> to auto-discover and attach to a running Chrome instance via CDP, inheriting existing logins, cookies, and extensions.
• ›Adds browser-use --profile <name> open <url> to launch directly into an existing named Chrome profile (e.g. 'Default', 'Profile 1').
• ›Adds browser-use --cdp-url <ws-url> open <url> to connect to any browser via an explicit CDP WebSocket URL.
• ›Adds browser-use --headed open <url> flag to run a visible browser window for debugging.
• ›Adds browser-use -s <session> open <url> for named, parallel browser sessions; browser-use sessions lists all active sessions; browser-use close --all terminates them.

GitHub Copilot CLI v1.0.11 adds monorepo skill discovery, per-session working directories, and merged extension hooks.

$ git clone --branch v1.0.11 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v1.0.11

$ /new 'Run the full test suite and summarise failures'

statusLine.command: ~/.local/bin/my-status ${PROJECT:-default}

• ›Adds ~/.agents/skills/ as a personal skill discovery directory, aligning with the VS Code GHCP4A extension default.
• ›Supports statusLine.command path expansion for ~ and environment variables such as $HOME and ${VAR:-default}.
• ›Custom instructions, MCP servers, skills, and agents are now discovered at every directory level from the working directory up to the git root, enabling full monorepo support.
• ›/new and /clear commands now accept an optional prompt to seed the new session with a first message.
• ›/clear now abandons the current session entirely, while /new starts a fresh conversation with the old session kept backgrounded.

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

Pi v0.62.0 adds extensible built-in tool rendering, unified sourceInfo provenance, and AWS Bedrock cost allocation tagging.

$ git clone --branch v0.62.0 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:
$ git checkout v0.62.0

• ›Enables extension authors to override rendering of built-in read/write/edit/bash/grep/find/ls tools with custom renderCall/renderResult components via extensible ToolDefinitions.
• ›Adds unified sourceInfo struct (path, scope, source metadata) to all resources, commands, tools, skills, and prompt templates — visible in autocomplete, RPC discovery, and SDK introspection.
• ›New requestMetadata option on BedrockOptions forwards key-value pairs to the Bedrock Converse API for AWS Cost Explorer split cost allocation tagging.

Qwen Code v0.13.0 adds hooks, agent arena, skills directories, /btw and /context commands, and system prompt customization.

$ git clone --branch v0.13.0 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.13.0

$ /btw What does the --dry-run flag do in this codebase?

$ /context

$ qwen auth

• ›Adds --system-prompt customization options to both the SDK and CLI for overriding the default system prompt.
• ›Adds /btw slash command for asking ephemeral side questions without affecting the main conversation context.
• ›Adds /context slash command to display a breakdown of context window token usage.
• ›Adds auth CLI command for authentication flows.
• ›Implements a hooks extension mechanism with 10 core event hooks covering session lifecycle and tool execution events.

Jan v0.7.9 adds NVIDIA NIM cloud provider, Spanish locale, expanded file attachments, and auto-scaling context length

$ git clone --branch v0.7.9 https://github.com/janhq/jan.git
# already have the repo? check out this version:
$ git checkout v0.7.9

• ›Adds setting to auto-increase ctx_len when the context window fills up, preventing truncation without manual tuning.
• ›Removes the 32k ctx_len enforcement that was previously applied on API server start, allowing models to use their native context limits.
• ›Adds NVIDIA NIM as a predefined cloud provider in the model hub.
• ›Adds a configurable default model for the local API server.
• ›Adds a default assistant setting, letting users pin a preferred assistant across sessions.

llama.cpp b8493 adds OpenCL q6_K GEMM and GEMV kernels for Adreno GPUs.

$ git clone --branch b8493 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8493

• ›Adds OpenCL q6_K GEMM (gemm_noshuffle_q6_k_f32) and GEMV kernels for Adreno GPUs, enabling quantized 6-bit inference acceleration on mobile/embedded hardware via the OpenCL backend.

$ git clone --branch b8477 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8477

Phoenix v13.17.0 adds an experimental GQL simulated bash tool for agents and one-click span ID copying in the UI.

$ git clone --branch arize-phoenix-v13.17.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v13.17.0

• ›Adds experimental px-gql simulated bash tool for agent workflows.
• ›Adds one-click copy of span IDs to clipboard in the trace UI.

Milvus v2.6.13 adds Google Gemini embedding support and a Go SDK TruncateCollection method.

$ git clone --branch v2.6.13 https://github.com/milvus-io/milvus.git
# already have the repo? check out this version:
$ git checkout v2.6.13

• ›Adds TruncateCollection method to the Go SDK client for clearing all data in a collection without dropping it.
• ›Adds Google Gemini as a built-in text embedding function, configurable via a Gemini API key, supporting models including Gemini Embedding 2.
• ›Adds query metrics for JSON-related filter expressions to improve observability of JSON field query performance.