The AI Toolchain — issue -335, September 15, 2025

Lightweight coding agent that runs in your terminal

Codex CLI v0.36.0 adds GPT-5-Codex, session resumption, JSON-RPC expansions, MCP interface, and experimental context compaction.

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

$ codex login --api-key "sk-...your-key..."

• ›Adds GPT-5-Codex model support: faster on easy tasks, harder on complex tasks, with improved code quality and better AGENTS.md steerability.
• ›New codex resume command lets practitioners pick up old conversations where they left off.
• ›New codex login --api-key command stores API credentials in CODEX_HOME/auth.json for explicit, persistent authentication.
• ›Adds initial MCP (Model Context Protocol) interface and documentation for integrating external tool servers.
• ›Expands JSON-RPC server with SetDefaultModel (passing None clears the default), UserInfo, and reasoning_effort field in NewConversationResponse.

The open source coding agent.

OpenCode v0.9.1 ships a new desktop/web app package.

$ git clone --branch v0.9.1 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v0.9.1

• ›Adds a desktop/web app package, enabling OpenCode to be run as a desktop or web application.

PydanticAI v1.0.7 adds MCP metadata filtering, FunctionToolset defaults, and improved RunContext prompt access.

$ git clone --branch v1.0.7 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:
$ git checkout v1.0.7

from pydantic_ai.toolsets import FunctionToolset

toolset = FunctionToolset(
    strict=True,
    sequential=False,
    requires_approval=True,
    metadata={"source": "internal"}
)

from pydantic_ai.tools import ToolDefinition

def only_safe_tools(tool_def: ToolDefinition) -> bool:
    meta = tool_def.metadata or {}
    return not meta.get("destructive", False)

• ›Adds ToolDefinition.metadata field to carry MCP metadata and annotations, enabling filtering of MCP tools by metadata.
• ›Adds support for default values for strict, sequential, requires_approval, and metadata parameters on FunctionToolset, reducing per-tool boilerplate.
• ›When a run starts with a message history ending in a ModelRequest, its content is now available in RunContext.prompt.
• ›Removes the requirement to install mcp or logfire extras when using Temporal or DBOS integrations.
• ›Combines consecutive AG-UI user and assistant messages into a single model request/response.

SRE Agent - CNCF Sandbox Project

HolmesGPT 0.14.1 adds a free-form Tempo toolset, LLM-based message summarization via a Transformer mechanism, and api_base/api_version Azure OpenAI config keys.

$ git clone --branch 0.14.1 https://github.com/HolmesGPT/holmesgpt.git
# already have the repo? check out this version:
$ git checkout 0.14.1

• ›Adds api_base and api_version config keys for Azure OpenAI support in the HolmesGPT configuration.
• ›Introduces LLM-based message summarization via a new Transformer mechanism to optimize context window handling, improving accuracy with metrics and trace data.
• ›Adds a free-form Tempo toolset enabling natural-language trace queries such as 'Why was there an increase in latency?' and 'Find the cause of recent HTTP 500 errors.'
• ›Implements an LLM Model Registry allowing context size overrides per model.
• ›Adds max_tokens and max output tokens fields to Holmes responses.

Phoenix v11.33.0 adds prompt labels, an experiment compare list page, repetitions carousel, and annotation popovers in experiment details.

$ git clone --branch arize-phoenix-v11.33.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v11.33.0

• ›Adds prompt labels to prompt versions, now visible in prompt details view.
• ›Releases a new experiment compare list page for side-by-side experiment comparison.
• ›Adds a repetitions carousel on the experiment compare grid page to browse repeated runs.
• ›Adds an annotation popover on the experiment details view for inline annotation inspection.
• ›Enables paging on the experiment compare details view to navigate large result sets.

Langfuse v3.111.0 extends PostHog integration to export non-trace scores alongside existing trace data.

$ git clone --branch v3.111.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.111.0

• ›Exports non-trace scores (e.g. session and span-level scores) via the PostHog integration, expanding beyond trace-only score forwarding.
• ›Includes ticket ID in support email subjects, making it easier to correlate in-app support requests with email threads.

Composio v0.8.12 adds connected account lookup by user ID

$ git clone --branch v0.8.12 https://github.com/ComposioHQ/composio.git
# already have the repo? check out this version:
$ git checkout v0.8.12

• ›Adds the ability to retrieve a connected account for a specific user via user_id.