Cognition AI 16小时前
Devin June '24 Product Update
index_new5.html
../../../zaker_core/zaker_tpl_static/wap/tpl_guoji1.html

 

本文介绍了Devin AI的最新更新,重点关注其提升的工程协作能力和用户控制。更新包括直接使用Devin的机器进行文件编辑和终端操作,以及通过Playbooks实现自动化工程任务。此外,还新增了知识库功能,方便用户分享文档和自定义库,提升Devin在特定环境中的表现。文章还提到了Devin与Slack、GitHub等平台的集成,以及新的秘密管理和会话状态控制功能,增强了用户体验和安全性。最后,文章强调了Devin在工程领域的应用前景,并提供了访问方式。

💻 **直接控制与交互:** 用户现在可以直接使用Devin的机器,通过VSCode进行文件编辑、终端操作,并利用交互式浏览器完成需要辅助的浏览器任务,比如处理CAPTCHAs。

🤖 **Playbooks实现自动化:** Playbooks允许用户定义Devin执行常见工程任务的步骤和上下文,类似于程序但没有严格的语法,支持快速指定任务步骤、成功标准和安全措施。Playbooks可在组织内共享,也可以从社区库中获取,从而减少指导Devin完成任务所需的时间。

📚 **知识库功能:** 新增知识库功能,用户可以分享文档、自定义库等信息,帮助Devin更好地适应特定组织环境。知识库与Playbooks结合使用,可以为Devin提供通用背景知识和特定任务的最佳实践。

🔗 **集成与控制增强:** Devin现在与Slack、GitHub、Linear等平台集成,支持事件驱动的工作流程,并提供Slack通知。新增的秘密管理器增强了安全性,用户可以安全地管理API密钥等敏感信息。用户还可以通过新的控制功能暂停、终止或保持会话运行,并改进了对Devin操作的可见性,包括全局工作差异和命令历史记录。

Use Devin’s Browser, Editor and Shell

It can sometimes be more convenient to directly take actions for Devin, rather than providing instructions for Devin to follow.

We’re excited to share that you can now directly use Devin’s machine. The new “Use Devin’s Machine” button in the web interface opens VSCode in a new tab. Using VSCode, you can directly read and edit Devin’s files, as well as open up a terminal in Devin’s machine.

Similarly with the browser, the new Interactive Browser lets you directly use Devin’s browser. This is especially helpful for browser tasks where Devin may require assistance, such as completing CAPTCHAs.

With Playbooks, Devin is better equipped for repetitive, multi-step engineering tasks

We’ve worked with a number of enterprises using Devin for tasks that are recurring in nature, such as adding unit tests or performing code migrations.

Playbooks are documents that provide Devin with the steps and context needed to complete common engineering tasks. They’re like programs, without the rigid syntax. With Playbooks, you can quickly specify not only the steps Devin needs to take to successfully complete a task, but also the success criteria and guardrails.

Playbooks can be easily shared within an organization, so once one engineer successfully coaches Devin to complete a task, other engineers can replicate that success with similar tasks.

We also launched a community gallery of Playbooks for tasks that are common across our users - e.g. full stack app development and deployment, web scraping, and Python notebook to script conversion. Playbooks minimize the back and forth needed to guide Devin to complete a task successfully, as they communicate the necessary instructions and details upfront.

Quickly equip Devin with necessary organizational context

Just as with onboarding a new engineer, onboarding Devin requires an initial investment in knowledge transfer.

With the new Knowledge feature, you can share documentation, tips, custom internal libraries, and other materials that Devin needs to be successful within your organization. Devin will use relevant Knowledge automatically to improve its performance in your environment.

We recommend using Knowledge in combination with Playbooks. Create Playbooks for common, recurring tasks, and make sure to include best practices for those tasks in the Playbook. Add Knowledge to teach Devin general context about your organization that is relevant for all runs, with and without Playbooks.

Persisting Devin’s dev environment

With Machine Snapshots, Devin is able to hit the ground running in every new session.

Machine Snapshots are ‘save’ states for Devin. After you take a snapshot, you can start from that machine state (with everything you’ve downloaded/installed) on any future Devin run. You can use Machine Snapshots to:

    Perform one-time installation and setup and start new sessions from that point, every timeResume from a previously ended session

For each snapshot, you can also set a list of startup commands that will be run at the beginning of every run. The commands will run in sequence. For example, with startup commands, you can have Devin cd your-repo, git pull and npm install at the beginning of every session.

Work with Devin via Slack, Github, or Linear

There are now new ways of working together with Devin.

First, Devin can now be triggered in an event-driven way. For example, Devin can get to work immediately when build fails, or when it’s assigned a Linear ticket.

Devin will also respond to Slack messages, and will keep you updated on progress via Slack threads. You can also turn on Slack or desktop notifications for sessions started via the web app.

With native GitHub integration, Devin can also quickly create or review pull requests and respond to comments on its PRs.

Secure secrets sharing and authentication

The new Secrets manager provides flexibility and security in managing personal and organization secrets, such as API keys, passwords, and authentication tokens. Define organization-level secrets with Devin once, and Devin will automatically retrieve relevant secrets as needed in future sessions. Secrets can alternatively be provided for a single session. Secrets are stored securely and revocable at any time.

You can also choose to share cookies with Devin, which enables Devin to login to websites without needing a password.

New controls for session state

Directly pause, terminate, or “keep alive” a session with Devin in the web interface. With “Keep alive”, sessions stay alive until users manually terminate them. This allows users to continue accessing any services the session may be running.

Improved visibility into Devin’s actions

It’s now faster and easier to verify Devin’s work. With Global Work Diff, users can pull up a summary of all code changes Devin has made up to any point in time. For more detail, users can also view the file edit histories for any individual file.

Finally, with Command History, users can see a summary of all commands Devin has run across all shells.

Hire Devin

We are working hard to scale up access to Devin, and we continue to do weekly invite releases. To start using Devin for engineering work, join the waitlist or get in touch at info@cognition.ai.

test

Fish AI Reader

Fish AI Reader

AI辅助创作,多种专业模板,深度分析,高质量内容生成。从观点提取到深度思考,FishAI为您提供全方位的创作支持。新版本引入自定义参数,让您的创作更加个性化和精准。

FishAI

FishAI

鱼阅,AI 时代的下一个智能信息助手,助你摆脱信息焦虑

联系邮箱 441953276@qq.com

相关标签

Devin AI 工程
相关文章