# OpenAI adds beta agent computer-use support to Python and Node SDKs

Official SDKs gain computer-use APIs, while Claude Code updates its agent configuration and policy controls. Retrospective brief for 19:00–19:59 UTC on 29 September 2026.

## Python and Node SDKs add computer use to beta agents

OpenAI Python 3.22.0 and Node 7.25.0 both add computer-use support to beta agents. The release notes confirm the SDK surface change without establishing broad account availability or a production stability guarantee. Builders integrating computer-interaction agents can now review these SDK versions and the corresponding beta API contracts. 

[openai/openai-python v3.22.0 release notes](https://github.com/openai/openai-python/releases/tag/v3.22.0) · [openai/openai-node v7.25.0 release notes](https://github.com/openai/openai-node/releases/tag/v7.25.0)

## Claude Code adds provider restrictions and plugin configuration commands

Claude Code 2.1.285 adds an allowedProviders managed setting to restrict which API providers a machine may use, an environment variable to disable WebFetch, and plugin configuration through a command or stdin. It also fixes switching models through the Agent SDK without updating output-token and compaction limits until restart, and fixes tools remaining available after an MCP server is disabled in SDK or non-interactive sessions. 

[anthropics/claude-code v2.1.285 release notes](https://github.com/anthropics/claude-code/releases/tag/v2.1.285)

## The Hexagon backend gains additional FP32 activation support

The llama.cpp b11260 development build adds FP32 GELU_ERF and GEGLU_ERF support to its Hexagon backend and reduces register pressure in the related kernels. These activation operations are part of model inference. This is a prerelease backend update; the notes do not claim that it makes every model compatible with Hexagon. 

[ggml-org/llama.cpp b11260 release notes](https://github.com/ggml-org/llama.cpp/releases/tag/b11260)