Acyclic
Runs large coding tasks as parallel agent swarms in sandboxes and merges the results into one patch or pull request.
At a Glance
Engagement
Available On
Alternatives
Listed Oct 2026
About Acyclic
Acyclic Labs, which describes itself as backed by Y Combinator, builds Acyclic, a service that splits a task across many coding agents, each in its own sandbox. The agents work in parallel and their output is merged into a single patch or pull request. It is operated through a CLI, a web dashboard, or MCP-connected coding agents.
What It Is
Acyclic is an agent-swarm runner for large agentic workflows. You describe a task, such as fixing a flaky test, and the swarm fans the work out across agents, lets you fork branches worth exploring, and keeps results you verified. The company says this makes long workflows finish exponentially faster.
How the Workflow Runs
After installing the acyclic CLI (via npm or an install script on macOS and Linux), you sign in through the browser, run acyclic swarm run with a task and --dir . to upload working files (no history or secrets), and optionally --watch to follow agents live. The merged result can be applied locally with acyclic swarm pull --apply or turned into a pull request from the dashboard.
Agent and Tool Integration
Guides cover using swarms as a tool for Claude Code, Codex, Cursor or custom harnesses, adding MCP servers so every agent gets extra tools, choosing the model each agent runs, using your own OpenAI API key, and letting agents start GPU clusters in your own Google Cloud project with gcloud and SkyPilot.
Current Status
Acyclic says it is onboarding teams gradually, starting with those already running agents in volume, such as teams running many Claude Code or Codex sessions on large codebases.
Community Discussions
Be the first to start a conversation about Acyclic
Share your experience with Acyclic, ask questions, or help others learn from your insights.
Pricing
Contact sales
Gradual team onboarding; teams request a conversation with the founders about scaling agent swarms.
- Startups building domain-specific agents that need to run at scale
- Companies already operating swarms of agents in production
- Teams running many Claude Code or Codex sessions on large codebases
Capabilities
Key Features
- Parallel agent swarms, each agent in its own sandbox
- Merges results into one patch or pull request
- CLI with login, run, watch and pull --apply commands
- Live watching of agents and branches
- Forking of promising branches
- Dashboard for starting and viewing swarms
- MCP server connections for extra agent tools
- Model selection for agents
- Bring your own OpenAI API key
- GCP and SkyPilot GPU cluster jobs
- llms.txt flow written for agents