Give the task a home.
Queue a worker through MCP with its task, instructions, timeout, and expected deliverables. Each worker runs in an isolated VM.
PUBLIC SOURCE / AGENT INFRASTRUCTURE
Give every agent
a workspace.
Keep the whole
team in view.
SwarmForge is a self-hosted orchestration platform for developers running parallel AI coding agents. Coordinate OpenCode workers in isolated Freestyle VMs through MCP, track task progress, and preserve their deliverables.
Self-hosted control plane · Linux x64 · MCP interface
TEAM / CHECKOUT FLOW03 workers
●
Implement checkoutcoder · isolated workspace
RUNNINGReading the existing payment flow…
◉
Review edge casesreviewer · follow-up ready
WAITING✓
Document the APIwriter · outputs preserved
COMPLETEOUTPUTS
review.md SHA-256 verified
change-summary.md preserved
Illustrative task view. See the CLI for your actual workers and results.
Preserved files survive VM cleanup.Preserve deliverables before VM cleanup.
START IN YOUR TERMINAL
One command. Your infrastructure.
Install the CLI, configure your providers, and check local readiness.
curl -fsSL https://getswarmforge.tech/install | bashRequires Linux x64 with glibc and configured VM/model providers. Inspect the script.
FROM ASSIGNMENT TO HANDOFF
A clear path through
parallel work.
SwarmForge coordinates multiple coding workers concurrently. Your MCP manager assigns tasks; OpenCode executes them in Freestyle VMs.
See work as it moves.
Inspect response activity, results, reported test outcomes, and measured token usage. Find workers with dashboard search and filters.
Keep what matters.
Preserve artifacts, download them with checksum verification, inspect Git handoffs, and clean up eligible retained VMs.
BUILT FOR THE OPERATOR
Control you can
actually use.
SwarmForge connects your existing tools through a durable control plane you run yourself.
Read the setup guideMCP-native coordination
Spawn workers, send follow-up messages, and inspect results from an MCP client.
Durable results and artifacts
Read bounded text directly. Save complete files through authenticated, verified downloads.
Visible usage and cleanup
Configure cost estimates and budget alerts. Preview retention policies before enabling automatic expiry.
Recovery with context
Inspect failure guidance and preservation state, then choose the appropriate next action.
UNDERSTAND THE SYSTEM
Parallel agents.
Explicit responsibilities.
SwarmForge manages environments, task state and durable handoffs. Your manager decomposes the job, reviews evidence and integrates source changes.
Learn the orchestration model
Understand manager, coordinator, worker and harness responsibilities, then compare local processes, worktrees and remote VMs.
Multi-agent coding orchestration guide →Explore the architecture
Follow the path from MCP assignment to isolated OpenCode workers, SQLite state and preserved results.
SwarmForge system architecture →Choose a workflow
See practical ways to divide implementation, manage dependencies and hand off bounded tasks.
Parallel development workflow →BUILD IN THE OPEN
Your next agent team
starts with a task.
Explore the code, try a real workflow, and help shape what comes next.