OpenComputer开源云端编程智能体发布
OpenComputer 被介绍为一款开源云端编程智能体,团队可通过其免费云服务使用,或自行部署到 AWS、GCP 或 Azure 账户。创始人称,智能体框架与沙箱运行在同一持久 VM 上,会话通过 Git worktree 隔离,并在每轮后备份会话文件,以便在其他 VM 上恢复。
OpenComputer 为团队提供了运行在自有云基础设施上的编程智能体方案,并支持持久 VM 和可恢复会话,但其性能与成本数据仍仅来自厂商自述。
Hi HN!
Utpal here, I'm one of the founders of OpenComputer.
OpenComputer is an open source cloud coding agent for teams that can run in your own cloud account (AWS, GCP, Azure).
We run the agent harness and the sandbox on the same durable VM (architecture in the docs [1]). First commit lands in 24s; the alternatives we benchmarked took 53-72s (benchmarks below [2]).
Two other choices:
- One VM per team, scalable horizontally. Sessions are isolated in git worktrees. - Session files are backed up to a repo after every turn, so any session can resume on another VM.
We noticed that past ~640 session-hours a month, one always-on 4 vCPU VM is cheaper than per-session E2B sandboxes, for example. So at scale, this setup would not only be faster than current SOTA coding agents, but also more cost-efficient.
Feel free to try via cloud for free [4] or self host. We would love any and all feedback, and happy to answer any questions! Full demo below [3].
[1]: https://docs.opencomputer.dev/getting-started/architecture [2]: https://opencomputer.dev/blog/cloud-agents-as-fast-as-local/ [3]: https://x.com/IgorZIJ/status/2121309785882591711 [4]: https://app.opencomputer.dev