Parallel agent swarms
Split a larger task across coding agents running in separate sandboxes, allowing independent work to proceed concurrently.
Acyclic Labs
Acyclic Labs runs software tasks across parallel coding agents in isolated sandboxes, combining results into reviewable patches or pull requests via CLI, dashboard, or MCP.
Agent swarms for software development
Acyclic Labs is a developer tool for running agent swarms on software tasks. It divides a task among coding agents, runs each agent in its own sandbox, and combines the work into a single patch or pull request.
Users can start a swarm from the CLI, the dashboard, or a coding agent connected through MCP. The workflow is designed for large codebase tasks and teams already running multiple agent sessions, with controls for watching progress, exploring branches, reviewing the merged result, and applying it to a checkout.
Features
Branch exploration
Fork promising lines of investigation and retain the results that have been verified, rather than relying on a single agent path.
CLI and dashboard control
Run swarms from the command line or dashboard, and stream active progress with the --watch option.
Unified code output
Merge agent work into one patch, apply it to a local checkout after review, or open a pull request from the dashboard.
Codex CLI harness
Choose OpenAI's Codex CLI as the agent harness with --harness codex, instead of the default harness.
MCP tool access
Register MCP servers and select them for a swarm so agents can call the connected tools. Web search is built in.
Use cases
Large codebase changes
Distribute a broad engineering task across multiple isolated agents, then review one merged patch instead of coordinating separate working directories manually.
Exploring implementation options
Run several possible approaches or investigations in parallel, follow their progress, and keep the branches that produce verified results.
Command-line development workflows
Use the CLI to provide a project directory or public repository, monitor execution, and apply the reviewed output back to a local checkout.
Agent-assisted development environments
Start swarms from the dashboard or connect Claude Code or Cursor through an MCP configuration when a terminal installation is not preferred.
Tool-augmented research and coding
Give agents selected MCP services, such as built-in web search or a configured external server, when a task requires additional tools.
Pros and Cons
Pros
- Runs multiple coding agents in parallel, each in an isolated sandbox.
- Produces a consolidated patch and supports review before applying changes.
- Offers CLI, dashboard, and MCP-based ways to start or control swarms.
- Supports MCP tools and an alternate Codex CLI harness.
Cons
- The documented CLI workflow is currently for macOS and Linux; Windows support is not stated.
- The product is onboarding teams gradually, so access or availability may depend on a conversation with the Acyclic team.
- The supplied evidence does not specify pricing, usage limits, or detailed security and data-retention policies.
FAQ
How do you install Acyclic Labs?
Install the CLI with npm install -g @acyclic-labs/cli on macOS or Linux, then run acyclic --version to verify the installation. A platform-detecting install script is also available and can install the binary to ~/.local/bin, with ACYCLIC_INSTALL_DIR available to override the location.
How does CLI authentication work?
Run acyclic swarm login --site https://acyclic.dev to approve the machine in a browser. The login token lasts 90 days and can be revoked from the dashboard.
What inputs can a swarm use?
A swarm can use project files supplied with --dir . or a public Git repository supplied with --repo. The --dir workflow uploads working files without history or secrets, and --watch streams progress while the swarm runs.
What does a swarm produce?
The agents' work is merged into one patch. You can review it with --patch FILE before using --apply to update your checkout, or open a pull request from the swarm page in the dashboard.
Can swarm agents use external tools?
Acyclic supports connecting MCP servers through Settings → Credentials and Settings → Tools. After selecting a configured server when starting a swarm, every agent can call its tools. Web search is built in; the guide also documents TinyFish as a preview host and notes that other servers may require contacting Acyclic.
Quick Facts
- Category
- Developer Tool
- Primary workflow
- Parallel coding-agent swarms
- Supported CLI platforms
- macOS and Linux
- Output
- Merged patch or dashboard pull request
- Access
- Gradual team onboarding; contact-based conversation
- Authentication
- Browser-approved device token lasting 90 days
Acyclic Labs Alternatives
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads in Firecracker micro-VMs with private networking and SDK, CLI, or MCP control.
GitBot
GitBot is a local browser workspace for reusable coding bots powered by Claude Code, Codex, or OpenCode, with persistent threads and configurable permissions.
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repos and verifies changes with compilers, debuggers, sanitizers, and tests.
hob
hob is an independent workspace for coding agents, with local control over sessions, terminals, history, routing, and follow-up work.
Manta AI
Manta AI is an autonomous web app testing tool that maps app behavior, catches regressions, and generates tests from a URL, no scripts or selectors needed.
SonOf
SonOf connects to your repo and PM tool, audits your codebase, and turns approved work into shipped tickets with senior engineering review.