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.
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.
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.
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/github, so check it out!
To run your own swarm (check out acyclic.dev for a full guide)
``` npm install -g @acyclic-labs/cli acyclic swarm run --task <prompt> --dir <files_to_upload> ```
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/dashboard or stream the results back to the cli using:
``` acyclic swarm pull <swarm-id> --apply ```
Check out some more research examples at acyclic.dev/evals, right now our biggest useful swarm is 236 agents for a code migration.