Acyclic Labs

Teams running coding agents like Claude Code or Codex on large codebases hit tasks that are too big for a single agent to handle efficiently. Running agents one at a time or in an unmanaged way makes large or exploratory tasks slow, and there is no easy way to fan work out, explore multiple branches, and keep only the verified results. Startups building domain-specific agents and companies already operating agent swarms in production need a way to scale this beyond ad hoc scripts or single-agent runs.

Acyclic Labs provides a CLI and dashboard that split a task across many coding agents, each running in its own sandbox, working in parallel before merging into one patch or pull request. Users install the CLI, sign in once (valid 90 days), then run a command describing the task and the working directory, with a --watch flag to follow agents live. The result can be pulled and applied directly to a local checkout or opened as a pull request from the dashboard. The product also supports forking promising branches of work and watching every branch as it runs, and can be used without installing anything by starting a swarm from the dashboard or connecting a coding agent over MCP.