OpenHands 1.7.0 リリース、会話表示とサンドボックスKVM対応を追加 OpenHands Releases 1.7.0 - 2026-05-01
- オープンソースのAI開発エージェント「OpenHands」が1.7.0を公開し、会話カードとヘッダーへのLLMモデル表示や、/dev/kvmをサンドボックスに渡せるSANDBOX_KVM_ENABLED変数を追加した。
- 実行環境とUIの改善で開発支援を強化する。
English summary
- Open-source AI coding agent OpenHands shipped 1.7.0, adding LLM model display on conversation cards and the header plus a SANDBOX_KVM_ENABLED variable to pass /dev/kvm into sandbox containers, improving its execution environment and UI.
オープンソースのAI開発エージェント「OpenHands」が、バージョン1.7.0をリリースした。OpenHandsはユーザーの指示を受けてコードの編集、コマンド実行、ブラウザ操作などを自律的に行う汎用エージェント基盤で、ソフトウェア開発タスクの自動化を目指すプロジェクトとして注目を集めている。
1.7.0は通常のマイナーアップデートに位置付けられ、エージェントの実行環境やツール統合、LLMプロバイダ対応の改善などが含まれているとみられる。OpenHandsはサンドボックス化されたコンテナ内でエージェントを動作させるアーキテクチャを採用しており、安全性と再現性を保ちながら多様なモデル(Claude、GPT、ローカルLLMなど)を切り替えて利用できる点が特徴だ。継続的なリリースを通じて、ベンチマークSWE-benchなどで示される自律コーディング性能の底上げが図られている。
背景として、OpenHandsは旧称「OpenDevin」として登場し、Cognitionの「Devin」に対するオープンソース版として急速にスター数を伸ばした経緯がある。現在はAnthropicのClaude Codeや、GitHub Copilot Workspace、Cursor Agentなど、商用・OSS問わず自律型コーディングエージェントの競争が激化している領域だ。OpenHandsはCLIだけでなくクラウド版やGitHub連携も整備されており、エンタープライズ用途でも採用が進む可能性がある。
オープンソースのAI開発エージェント「OpenHands」が1.7.0を公開し、会話カードとヘッダーへのLLMモデル表示や、/dev/kvmをサンドボックスに渡せるSANDBOX_KVM_ENABLED変数を追加した。
リリースノートの詳細はGitHubのタグページで確認できる。利用中のユーザーは依存関係やプロンプト挙動の変化に留意したうえで、最新版への更新を検討するとよいだろう。
OpenHands, the open-source autonomous AI development agent formerly known as OpenDevin, has released version 1.7.0. The project provides a general-purpose agent framework capable of editing code, executing shell commands, and operating a browser on behalf of the user, and it has emerged as one of the more visible attempts to automate end-to-end software engineering workflows.
Version 1.7.0 appears to be a routine minor release, likely bundling refinements to the agent runtime, tool integrations, and LLM provider support rather than introducing sweeping architectural changes. While the precise contents are documented on the GitHub release page, the cadence of OpenHands releases has consistently focused on incremental improvements to reliability, prompt behavior, and the breadth of supported models and execution environments.
Architecturally, OpenHands runs its agents inside sandboxed containers, an approach intended to preserve safety and reproducibility while allowing users to swap between frontier models such as Anthropic's Claude and OpenAI's GPT family, as well as locally hosted open-weight LLMs. This model-agnostic design has been a core selling point for teams that want to evaluate different backends against the same agent loop, or that need to keep sensitive code on-premises. Continued iteration through releases like 1.7.0 is also tied to the project's effort to push autonomous coding performance higher on benchmarks such as SWE-bench, where OpenHands has historically posted competitive scores among open-source systems.
The project's trajectory is worth recalling for context. Launched as OpenDevin in early 2024 in response to Cognition Labs' closed Devin demo, it rapidly accumulated GitHub stars and contributors, and was later rebranded to OpenHands as the maintainers formalized governance around the All Hands AI organization. Since then, the autonomous coding agent space has become increasingly crowded, with Anthropic's Claude Code, GitHub Copilot Workspace, Cursor's agent mode, Devin itself, and a growing roster of open-source competitors such as Aider and SWE-agent all vying for developer mindshare.
Against that backdrop, OpenHands has been broadening beyond its original CLI roots. The team offers a hosted cloud version and tighter GitHub integration, including the ability to dispatch agents directly from issues and pull requests, which positions the project for potential enterprise adoption alongside its developer-tool persona. Whether that translates into meaningful commercial traction will depend in part on how reliably the agent can complete non-trivial engineering tasks without human babysitting, an area where every contender in the category is still visibly maturing.
For existing users, upgrading to 1.7.0 should be straightforward, but the usual caveats for agent frameworks apply. Minor releases in this ecosystem can shift prompt templates, tool definitions, or default model parameters in ways that subtly alter agent behavior, and dependency updates may interact with custom sandbox configurations or self-hosted LLM endpoints. Reviewing the published release notes on GitHub before rolling the update into production pipelines is advisable, particularly for teams that have built workflows around specific agent outputs.
More broadly, the steady release rhythm from OpenHands underscores how quickly the autonomous coding agent category is evolving. As proprietary offerings race to demonstrate end-to-end task completion and open-source projects iterate in public, the gap between commercial and community implementations remains narrow enough that incremental releases like 1.7.0 are worth tracking for anyone evaluating where to place long-term bets in AI-assisted software development.
本ページの本文・要約は AI による自動生成です。正確性は元記事 (github.com) をご確認ください。