The AI Toolchain — issue -027, July 23, 2026

Agno v2.8.1 adds Marengo video embeddings, Slack peer-agent comms flag, and a loop-guard for Learning Stores.

$ git clone --branch v2.8.1 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:
$ git checkout v2.8.1

slack_agent = Agent(
    tools=[SlackTools(respond_to_other_agents=True)],
    ...
)

learning_store = LearningStore(
    extraction_tool_call_limit=5,
    ...
)

• ›Adds respond_to_other_agents flag to the Slack integration to enable peer-agent communication between Slack-connected agents.
• ›Adds extraction_tool_call_limit to Learning Stores to cap runaway tool calls and prevent infinite loops.
• ›Adds stream_sub_agent_events support across all Context Providers.
• ›Adds Marengo video embeddings support to TwelveLabsTools.

langchain-openai 1.4.1 adds LangSmith gateway support via environment variable.

$ git clone --branch langchain-openai==1.4.1 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain-openai==1.4.1

• ›Supports routing OpenAI calls through the LangSmith gateway, configurable via an environment variable.

$ git clone --branch langchain-anthropic==1.5.1 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain-anthropic==1.5.1

$ git clone --branch langchain-core==1.5.1 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain-core==1.5.1

PydanticAI v2.16.0 adds ToolFailed, Model Armor, run_id support, Mistral caching, and OpenAI moderation surface.

$ git clone --branch v2.16.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:
$ git checkout v2.16.0

from pydantic_ai import ToolFailed

@agent.tool
async def lookup_user(ctx, user_id: str) -> str:
    if not user_id.startswith('u_'):
        raise ToolFailed('user_id must start with u_; got: ' + user_id)
    return fetch_user(user_id)

result = await agent.run(
    'Summarize the threat landscape',
    model_settings={
        'mistral_prompt_cache_key': 'threat-landscape-v1',
        'parallel_tool_calls': True,
    },
)

result = await agent.run(
    'Analyze this incident report',
    run_id='incident-2025-07-14-001',
)

• ›Adds mistral_prompt_cache_key setting and passes parallel_tool_calls to the Mistral SDK via model settings.
• ›Adds openai_moderation to OpenAIChatModelSettings and exposes Chat Completions moderation results in provider_details.
• ›Adds Google Model Armor support for Google Cloud via GoogleModelSettings.
• ›Adds optional run_id= parameter to agent runs, durable wrappers, and UI adapters for correlating runs.
• ›Adds ToolFailed exception class for surfacing model-visible tool failures without triggering retries.

GitHub Copilot CLI v1.0.74 adds Open Plugin Spec v1 / mcp.json support, gemini-3.6-flash, and a dedicated plan-mode model picker.

$ git clone --branch v1.0.74 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v1.0.74

$ /model plan gemini-3.6-flash

$ /model plan off

• ›Adds /model plan (or /model --plan) subcommand to select a dedicated model for plan mode; accepts a model ID, off to clear the override, or no argument to open the interactive picker — reverts to the session model on exit.
• ›Adds support for Open Plugin Spec v1 plugin manifests and mcp.json configuration files.
• ›Adds gemini-3.6-flash as a supported model.
• ›The /mcp add and /mcp edit wizard now preserves = characters in environment variable values (e.g. base64-padded secrets and tokens).
• ›The $ interactive shell shortcut now opens a shell even while the agent is working.

an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM

Goose v1.44.0 adds --edit session flag, overlapping-window command classifier, new providers (Sakana AI, Fireworks AI, iFlytek Spark, OllamaCloud), and per-message usage stats.

$ git clone --branch v1.44.0 https://github.com/aaif-goose/goose.git
# already have the repo? check out this version:
$ git checkout v1.44.0

$ goose session --edit <session-id>

• ›Adds --edit session flag to edit conversation history before forking a session.
• ›Chunks command-classifier input with overlapping windows to improve security classification coverage.
• ›Adds OllamaCloudProvider with dynamic model discovery and automatic context-limit detection.
• ›Adds declarative Sakana AI provider for the OpenAI-compatible Fugu API.
• ›Adds Fireworks AI as a declarative provider.

$ git clone --branch v1.44.0 https://github.com/block/goose.git
# already have the repo? check out this version:
$ git checkout v1.44.0

$ goose session --edit

Lightweight coding agent that runs in your terminal

Codex CLI adds supports_standalone_web_search setting so custom model providers can opt into the standalone web.run tool.

$ git clone --branch rust-v0.146.0-alpha.3.1 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.146.0-alpha.3.1

• ›Adds supports_standalone_web_search model-provider setting (defaults to false) that lets custom Responses API providers opt into the standalone web.run web-search tool when web search is enabled.
• ›Routes standalone web search requests through the custom provider's own endpoint and authentication when supports_standalone_web_search is enabled.

$ git clone --branch rust-v0.146.0-alpha.5 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.146.0-alpha.5

$ curl -fsSL https://chatgpt.com/codex/install.sh | CODEX_INSTALLER_USE_RELEASES_OPENAI_COM=false sh

Zed v1.12.0 adds Git staged/unstaged grouping, multi-select finders, adaptive Anthropic thinking, and format-on-save by changed lines.

$ git clone --branch v1.12.0 https://github.com/zed-industries/zed.git
# already have the repo? check out this version:
$ git checkout v1.12.0

{
  "format_on_save": "modifications"
}

{
  "language_models": {
    "anthropic": {
      "custom_models": [
        {
          "name": "claude-custom",
          "model": "claude-opus-4-5",
          "mode": { "type": "adaptive" }
        }
      ]
    }
  }
}

{
  "lsp_results_location": "pane"
}

• ›Adds modifications and modifications_if_available options to format_on_save to format only Git-changed lines instead of the entire file; also scopes remove_trailing_whitespace_on_save and ensure_final_newline_on_save to changed lines in these modes.
• ›Adds lsp_results_location global setting and per-action open_results_in option to control filterable result pickers with previews for editor: find all references, editor: go to definition, and editor: go to implementation.
• ›Adds support for adaptive thinking in custom Anthropic models via the "mode": { "type": "adaptive" } configuration key.
• ›Adds the supports_fast_mode setting for enabling fast mode on custom Anthropic models.
• ›Adds GPT-5.6 Sol, Terra, and Luna models to the Amazon Bedrock provider via the bedrock-mantle endpoint.

Jan v0.8.4 adds an OpenAI-compatible gateway, Responses API, native web search, and migrates credentials to the OS keyring.

$ git clone --branch v0.8.4 https://github.com/janhq/jan.git
# already have the repo? check out this version:
$ git checkout v0.8.4

• ›Adds an OpenAI-compatible translating gateway, OpenAI Responses API support, and unified reasoning across providers.
• ›Moves provider settings and API keys from localStorage to a backend-managed store, with secrets written to the OS keyring — existing data is migrated automatically on first launch.
• ›Adds native web_search/web_fetch capability via tauri-plugin-websearch.
• ›Adds per-model chat template kwargs for llama.cpp models.
• ›Adds a toggle for folding interim text into the reasoning trace in chat.

CUDA NVFP4 W4A4 activation quantization improved with fused kernels and intrinsics for faster inference on NVIDIA GPUs.

$ git clone --branch b10099 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10099

• ›Improves NVFP4 W4A4 activation quantization on CUDA with fused per-channel amax and quantization kernels, 32-byte loads, and nvfp4x4 intrinsic support where available, reducing overhead during quantized matrix multiplication.

$ git clone --branch b10094 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10094

$ llama-server -hf repo:Q3_K_M -hfd repo:Q8_0

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

Ollama v0.32.3 adds CUDA on Windows ARM64, B200 GPU support, Laguna 2.1 model capabilities, and restored Claude Code Channels.

$ git clone --branch v0.32.3 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.32.3

• ›Adds CUDA support on Windows ARM64, enabling GPU-accelerated inference on ARM-based Windows devices.
• ›Adds B200 GPU support via CUDA 12 (compute capability 10.0 on Linux).
• ›Reduces memory use on Linux CUDA and ROCm iGPUs through Direct I/O (dio) enablement.
• ›Adds chat, thinking, and tool calling support for Laguna 2.1 models, including a Metal inference fix.
• ›Updates the MLX and llama.cpp engines.

Arize Phoenix Evals 3.3.0 adds a toxicity gallery template with an input/output-agnostic benchmark.

$ git clone --branch arize-phoenix-evals-v3.3.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-evals-v3.3.0

• ›Adds a toxicity gallery template with an input/output-agnostic benchmark for evaluating toxic content across varied prompt and response surfaces.

Langfuse v3.224.1 rejects dataset-run-item writes in events-only mode and exposes uploaded media bytes via OpenTelemetry.

$ git clone --branch v3.224.1 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.224.1

• ›The API now rejects dataset-run-item surface writes when the server is running in events_only mode, preventing silent data loss.
• ›OpenTelemetry integration now exposes uploaded media bytes, making media payload sizes visible in OTEL telemetry.