原文整理页

开发者分享了在企业内部部署 OpenClaw AI 代理框架的实战经验,涵盖了从运维诊断到安全审计的多个应用场景

来源作者:Peter Steinberger 🦞 (@steipete)原始来源:https://x.com/steipete/status/2049015421236384185

中文导读

开发者分享了在企业内部部署 OpenClaw AI 代理框架的实战经验,涵盖了从运维诊断到安全审计的多个应用场景。

正文 Markdown

I’ve been running an “enterprise” openclaw install with 300 users @steipete (gcp, gke, gvisor sandbox, read only workspaces, etc) connected (read only) to new relic, github, google (with its own restricted account) and notion. It has changed how many teams work. Ops has been diagnosing customer issues in record time. It grabs logs from new relic, checks business contracts in notion and checks the repo to find where the transaction failed and why. “This customer support agent made this mistake here”. SecOps built a skill that does a RFC review (change management) and helps engineers address security issues before submitting the RFC. Devops is creating a similar one with infrastructure contracts in mind. We were recently at cloud next as a team checking out things, we added an agent to the channel to help coordinate the offsite. It helped coordinate restaurants, bars, activities, it ran polls, selected places, provided everything we needed in terms of directions and bookings. We just enjoyed the con and would get random Slack pings “choose dinner options close by to Mandalay!”. Some things are a bit… hacky still that I’ve had to put together to reconcile workspaces of users but so far I’ve automated the deployment to GKE and updating the PVC and even rotating secrets with SecretManager (injecting them at pod startup). The keeping it away from PII, read only to prod logs, sandboxing execs and giving it very reduced scope does help some of the anxiety of it doing catastrophic things It certainly is fascinating opening up all the user sessions as an admin and seeing all their use cases in every department. How the engineers use it, HR, Ops, etc.