The AI Toolchain — issue -139, April 2, 2026

Agno v2.5.14 adds fallback model chains for Agents and Teams, SAS token auth for Azure Blob, and a Slack workspace search tool.

$ git clone --branch v2.5.14 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:
$ git checkout v2.5.14

agent = Agent(
    model=OpenAIChat(id="gpt-4o", base_url="http://localhost:1/v1", retries=0),
    fallback_models=[Claude(id="claude-sonnet-4-20250514")],
)

• ›Adds fallback_models parameter to Agent and Team constructors, letting you specify an ordered list of backup models (e.g. Claude) that are tried automatically when the primary model fails.
• ›Adds SAS token authentication support to AzureBlobConfig for Azure Blob Storage connections.
• ›Adds a workspace search tool to SlackTools.

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

CrewAI 1.13.0 adds unified runtime state serialization, richer telemetry spans, token usage events, and an A2UI extension.

$ git clone --branch 1.13.0 https://github.com/crewAIInc/crewAI.git
# already have the repo? check out this version:
$ git checkout 1.13.0

from crewai.events import LLMCallCompletedEvent
from crewai.event_bus import event_bus

@event_bus.on(LLMCallCompletedEvent)
def on_llm_done(event: LLMCallCompletedEvent):
    print(event.token_usage)

• ›Adds RuntimeState RootModel for unified state serialization across crew runs.
• ›Enhances the event listener with new telemetry spans covering skill and memory events.
• ›Adds A2UI extension with v0.8/v0.9 support, schemas, and documentation.
• ›Emits token usage data in LLMCallCompletedEvent for per-call cost and usage tracking.
• ›Reduces framework overhead via a lazy event bus and skips tracing entirely when disabled.

Your own personal AI assistant. Any OS.

OpenClaw v2026.4.2 adds Task Flow orchestration, a before_agent_reply plugin hook, Android assistant entrypoints, and new Feishu/Matrix collaboration features.

$ git clone --branch v2026.4.2 https://github.com/openclaw/openclaw.git
# already have the repo? check out this version:
$ git checkout v2026.4.2

$ openclaw flows

# In your openclaw config
agents:
  defaults:
    compaction:
      notifyUser: false

# In your openclaw config
plugins:
  entries:
    diffs:
      config:
        viewerBaseUrl: "https://diffs.example.com"

• ›Adds openclaw flows inspection and recovery primitives for durable Task Flow state and revision tracking, with managed-vs-mirrored sync modes for persistent background orchestration.
• ›Adds managed child task spawning with sticky cancel intent, letting external orchestrators stop scheduling immediately and settle parent Task Flows to cancelled once active child tasks finish.
• ›Adds api.runtime.taskFlow plugin seam so plugins and trusted authoring layers can create and drive managed Task Flows from host-resolved context without passing owner identifiers on each call.
• ›Adds before_agent_reply plugin hook so plugins can short-circuit the LLM with synthetic replies after inline actions.
• ›Adds Android assistant-role entrypoints and Google Assistant App Actions metadata so OpenClaw can be launched from the assistant trigger and receive prompts into the chat composer.

PydanticAI v1.76.0 adds agent self-reference in RunContext and automatic image-generation fallback via subagent.

$ git clone --branch v1.76.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:
$ git checkout v1.76.0

@agent.tool
async def my_tool(ctx: RunContext[MyDeps]) -> str:
    current_agent = ctx.agent  # newly available in v1.76.0
    return f"Running as: {current_agent.name}"

• ›Adds agent attribute to RunContext, giving tools and callbacks direct access to the running agent instance.
• ›Adds automatic fallback for ImageGeneration: when the main model lacks image-generation capability, PydanticAI transparently delegates to a subagent running a dedicated imagegen model.
• ›Updates the Mistral integration to support mistralai SDK v2.

browser-use 0.12.6 adds CDP navigation timeout, model pricing URL override, and an install-lite CLI script.

$ git clone --branch 0.12.6 https://github.com/browser-use/browser-use.git
# already have the repo? check out this version:
$ git checkout 0.12.6

• ›Adds --connect/--cdp-url CDP navigation timeout support to prevent hung sessions when connecting to external browsers.
• ›Adds ability to override the model pricing URL via the tokens module (feat(tokens): allow overriding model pricing URL).
• ›Adds install-lite.sh script for a lighter-weight CLI installation path.
• ›Adds option to disable SignalHandler so host applications retain control of signal handling.

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 v2.1.91 adds MCP large-result passthrough, plugin executables, and multi-line deep link support.

$ git clone --branch v2.1.91 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:
$ git checkout v2.1.91

{ "_meta": { "anthropic/maxResultSizeChars": 500000 }, "content": "<full schema here>" }

$ open 'claude-cli://open?q=Summarize%20this%20file%0AHere%20is%20the%20context'

# In your Claude Code settings.json
{ "disableSkillShellExecution": true }

• ›Adds MCP tool result persistence override via _meta["anthropic/maxResultSizeChars"] annotation (up to 500K), allowing large results like DB schemas to pass through without truncation.
• ›New disableSkillShellExecution setting to disable inline shell execution in skills, custom slash commands, and plugin commands.
• ›Supports multi-line prompts in claude-cli://open?q= deep links — encoded newlines (%0A) are no longer rejected.
• ›Plugins can now ship executables under bin/ and invoke them as bare commands from the Bash tool.

Glamourous agentic coding for all

Crush v0.55.0 adds a built-in self-configuration skill, skill disabling, and an experimental server-client architecture.

$ git clone --branch v0.55.0 https://github.com/charmbracelet/crush.git
# already have the repo? check out this version:
$ git checkout v0.55.0

{
  "$schema": "https://charm.land/crush.json",
  "options": {
    "disabled_skills": ["crush-config"]
  }
}

$ CRUSH_CLIENT_SERVER=1 crush

• ›Adds built-in 'Crush config' skill so Crush can configure itself — add providers, MCPs, LSPs, or change settings via natural language without manual JSON editing.
• ›Supports disabling any skill (built-in, global, or project) via the new disabled_skills array in crush.json.
• ›Introduces experimental RPC server-client architecture, enabled with the CRUSH_CLIENT_SERVER=1 environment variable.
• ›Serves generated Swagger API docs from the server when running in server-client mode.

GitHub Copilot CLI v1.0.16 adds a PermissionRequest hook for scripted tool approval and MCP timeline visibility.

$ git clone --branch v1.0.16 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v1.0.16

• ›Adds PermissionRequest hook to allow scripts to programmatically approve or deny tool permission requests.
• ›MCP tool calls now display tool name and parameter summary in the timeline.
• ›MCP servers now reconnect correctly with valid authentication when the working directory changes, and load correctly after login, user switch, and /mcp reload.

LocalAI 4.1 adds distributed clustering, built-in auth/quotas, local-ai agent run, fine-tuning, and min_p inference support.

$ git clone --branch v4.1.0 https://github.com/mudler/LocalAI.git
# already have the repo? check out this version:
$ git checkout v4.1.0

$ local-ai agent run --prompt 'Summarize the top CVEs from this week'

$ local-ai agent run pool.json

• ›New local-ai agent run CLI subcommand runs standalone agents from the terminal, with single-turn --prompt mode and pool-based configurations loaded from pool.json.
• ›Adds HF_ENDPOINT environment variable support so the downloader rewrites HuggingFace URIs to a corporate mirror or custom endpoint.
• ›Adds min_p sampling parameter support wired through all inference endpoints.
• ›Adds distributed cluster mode with smart VRAM-aware routing, node groups for workload isolation, min/max autoscaling, and drain/resume lifecycle management via API call.
• ›Adds built-in multi-user platform with OIDC/OAuth SSO, invite-only registration, per-user API keys, admin impersonation, and per-user quota enforcement.

llama.cpp b8639 adds vectorized flash attention for the WebGPU backend, with q4/q8 quantization support.

$ git clone --branch b8639 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8639

• ›Adds vectorized flash attention to the WebGPU (ggml-webgpu) backend, enabled when Q->ne[1] < 20, Q->ne[0] % 32 == 0, V->ne[0] % 4 == 0, and K->type == f16; also supports q4 and q8 quantized key/value types.
• ›Introduces a fast single-workgroup (nwg=1) path for vectorized flash attention that skips temporary/reduce staging buffers, reducing overhead for small batch sizes.
• ›Uses packed f16 K loads in the flash_attn_vec_split.wgsl shader for improved memory throughput on the WebGPU backend.

$ git clone --branch b8634 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8634

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

Ollama v0.20.0 adds Google Gemma 4 model family support across four sizes including MoE and dense variants.

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

$ ollama run gemma4:e2b

$ ollama run gemma4:26b

• ›Supports Google Gemma 4 models in four sizes: E2B, E4B, 26B (Mixture of Experts with 4B active parameters), and 31B (Dense).
• ›Adds SentencePiece-style BPE tokenizer support.

Arize Phoenix v13.23.0 bundles TypeScript SDK docs directly into npm packages.

$ git clone --branch arize-phoenix-v13.23.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v13.23.0

• ›Bundles TypeScript SDK documentation into npm packages for offline/local access.