Orchestrate AI agents
on your infrastructure
Break down complex development tasks across specialized AI agents. Review and merge through a structured pipeline. Your source code never leaves your servers.
Built with technologies you trust
AI coding tools are powerful alone.
Real development isn't.
Current AI tools throw a single agent at complex tasks and hope for the best. That's not engineering.
Single-agent chaos
One AI trying to do everything. No structure, no specialization, no reliability on complex multi-file tasks.
Code leaves your servers
Your source code, API keys, and proprietary logic sent to third-party infrastructure you don't control.
Black box execution
No visibility into what the AI did, why it made decisions, or what files it changed. You just get a diff.
Hope-driven development
No tests before implementation. No review before merge. No pipeline. Just paste the prompt and hope.
How it works
Three steps. Full control.
Decompose
Describe what you need built. Switchyard's Yardmaster analyzes the task and breaks it into discrete subtasks with dependencies, file scopes, and test requirements. Each task is scoped to exactly one concern.
Orchestrate
Specialized agents work in parallel on isolated git worktrees. Coders implement features, testers write tests, reviewers check quality, security analysts audit for vulnerabilities. Each agent has a specific role in the pipeline.
Review & Merge
Every change goes through your review pipeline. Approve, request changes, add context, or redirect blocked agents. You see full diffs, execution traces, and test results. Nothing ships without your sign-off.
Features
Everything you need to orchestrate AI agents
Source Code Sovereignty
Core architectureYour source code never touches our servers. The Switchyard Relay runs on your infrastructure. API keys are stored locally with AES-256-GCM encryption. The control plane sees only metadata: task status, step counts, token usage, and costs. Never file contents. Never code.
Control plane never sees: source_code, api_keys, file_contents, trace_data
Configurable Pipeline
Define your workflow per project. Set model tiers, timeouts, and custom instructions for each stage.
Human Review Queue
Context-aware review actions for every situation. Approve merges, request changes, provide guidance to stuck agents, or escalate security findings. Full diff and trace visibility.
Live Agent Viewer
Watch agents work in real-time. See every file edit, command execution, and decision as it happens. Complete execution trace for audit and debugging.
Bring Your Own Keys
Use your own API keys with any supported provider. Control your AI costs directly. Switch models or providers without changing your pipeline configuration.
Security
Your code never leaves your servers
This isn't a policy — it's the architecture. The control plane physically cannot access your source code.
Control Plane
switchyard.cloud- ✓ Dashboard & task board
- ✓ Pipeline orchestration
- ✓ Analytics & cost tracking
- ✓ Team management
Receives only metadata:
status, duration, token_count, cost
Your Infrastructure
your servers- ◆ Source code & repositories
- ◆ API keys (AES-256-GCM encrypted)
- ◆ Agent containers & worktrees
- ◆ Full execution traces
Everything stays here.
Nothing leaves. Ever.
"They run your code on their servers. We run on yours."
Pricing
Simple, predictable pricing
Platform fee covers orchestration. You control AI costs with your own API keys.
Pro
For individual developers and small teams
- ✓ 5 projects
- ✓ Unlimited agents
- ✓ Full pipeline configuration
- ✓ Analytics dashboard
- ✓ Email support
Team
For growing engineering teams
- ✓ 50 projects
- ✓ Unlimited members
- ✓ SAML SSO
- ✓ Immutable audit log
- ✓ Per-project pipeline config
- ✓ Role-based access control
- ✓ Priority support
Getting started
One command to deploy
Install the relay on any Linux box with Docker. Takes about 60 seconds.
Ready to orchestrate?
Deploy your relay, connect your repos, and let a team of AI agents work for you — on your terms, on your infrastructure.