Original Summary

Hi HN, we&#x27;re the co-founders of Acyclic (YC F26).<p>Acyclic runs any agentic workflow as a swarm. You give it a task and your project; it splits the work across many coding agents, each in its own sandbox, runs them in parallel, and merges what they produce into one patch or pull request.<p>Our thesis: agentic workflows should be realtime, no matter how complex the task. The bottleneck should be human ideation and design, whilst building happens at the speed of thought.<p>To bring this to you we make our agents recursively fork and speculative execute. This means that if an agent has 2 independent tasks to do, it will just split into 2 agents to solve both tasks quickly and merge back up the tree. Additionally, for sequential operations we have “speculative” agents which guess the next steps of the workflow as a pre-fetch step.<p>To make this efficient we are solving the infrastructure bottlenecks. We are solving the memory, communication, routing and durable execution issues. This infra is open source at acyclic.dev&#x2F;github, so check it out!<p>To run your own swarm (check out acyclic.dev for a full guide)<p>`` npm install -g @acyclic-labs&#x2F;cli acyclic swarm run --task &lt;prompt&gt; --dir &lt;files_to_upload&gt; `<p>A good (non-coding) first task to try out: Do deep research into a person. “Do deep research into Jeff Goldblum“ You can view the results at acyclic.dev&#x2F;dashboard or stream the results back to the cli using:<p>` acyclic swarm pull &lt;swarm-id&gt; --apply ``<p>Check out some more research examples at acyclic.dev&#x2F;evals, right now our biggest useful swarm is 236 agents for a code migration.

  • 情报分类：技术学习与提效
  • 分类依据：内容涉及技术、AI、软件工具或工程实践
  • 信息来源：Hacker News 新项目
  • 发布时间：2026/10/19 22:46:04