Most AI companies guard their tools like state secrets. We put ours on GitHub.
Three projects, thousands of hours of development, completely free. Here's why that's not crazy—it's strategic.
The Tools We've Released
Sygaldry: Modular AI Engineering
GitHub: greyhaven-ai/sygaldry
Our framework for building production AI applications. Think of it as LEGO blocks for AI systems:
- Pre-built agents for common tasks
- Provider-agnostic (OpenAI, Anthropic, Google, Mistral)
- Standardized evaluation frameworks
- Production-ready observability
Why we built it: We were copying the same patterns across every client project. That's insane.
Why we open sourced it: The best frameworks come from real-world use. Community feedback makes it better for everyone, including our clients.
ContextFrame: LLM Context Management
GitHub: autocontext/contextframe
Managing context for LLMs is a nightmare. ContextFrame makes it boring:
- Standardized schema for document storage
- Built on Lance columnar format (fast)
- Multimodal support out of the box
- Relationship tracking between documents
The problem it solves: Every AI application needs context management. Everyone builds it differently. Nobody shares solutions. Waste everywhere.
Real impact: We have seen context retrieval time reduced by 80% using ContextFrame.
Claude Code Config: AI Assistant Configuration
GitHub: greyhaven-ai/claude-code-config
Configure Claude Code (or any AI assistant) for specific workflows:
- 20+ workflow presets
- 20+ specialized agents
- 40+ automation hooks
- Team-wide standardization
Why this matters: AI agents are powerful but inconsistent. Configuration brings predictability.
The Business Case for Open Source
1. Recruitment Magnet
Good engineers want to see your code before joining. Our repos are our portfolio:
- Senior prospective hires mention our open source work
- Partnership conversations started from GitHub
- Countless inbound inquiries
Recruiting cost: $0
2. Client Trust Builder
"Can we see examples of your work?"
Instead of sanitized case studies, we point to GitHub. Real code. Real documentation. Real community usage.
Trust established in minutes, not months.
3. Industry Standards Setting
When your tools become the standard, you win by default:
- Sygaldry patterns appearing in other projects
- ContextFrame schema being adopted
- Claude Config becoming the de facto configuration approach for many dev teams
We don't need to sell our approach. The community does it for us.
4. Quality Forcing Function
Open source code can't hide:
- Bad architecture gets called out
- Security issues get reported
- Performance problems get benchmarked
This pressure makes us better engineers. Our client work benefits.
What We Don't Open Source
Let's be clear—we're not giving away the farm:
Client Solutions: Custom implementations stay private Business Logic: Industry-specific algorithms remain proprietary Data Models: Trained models and datasets are confidential Secret Sauce: Some innovations provide competitive advantage
The framework is open. The implementation is not.
Lessons from Two Years of Open Source
1. Documentation Is Everything
Code without docs is worthless. We spend 30% of development time on documentation. It pays off:
- Fewer support requests
- More contributions
- Better adoption
2. Community > Features
A small, engaged community beats thousands of passive users:
- Our active contributors provide better feedback than 5,000 stars
- Real users find real bugs
- Production use cases drive development
3. Maintenance Is Forever
Open source is a commitment, not a campaign:
- Weekly security updates
- Monthly feature releases
- Constant community engagement
Budget 20% of engineering time for maintenance. Minimum.
How to Use Our Tools
For Startups
Start with Sygaldry. It's opinionated but flexible:
from sygaldry import Agent, Tool
agent = Agent("customer-support")
agent.add_tool(Tool.from_preset("email-responder"))
response = agent.run(customer_query)
Production-ready AI in 5 lines.
For Enterprises
ContextFrame for document management:
from contextframe import Dataset
ds = Dataset("enterprise-knowledge")
ds.add_documents(your_docs)
ds.build_embeddings()
context = ds.search(query)
Compliant, scalable, boring. Perfect for enterprise.
For Teams
Claude Config for standardization:
claude-config init --preset team-backend
claude-config apply --global
Everyone codes the same way. Onboarding time cut by 50%.
The Open Source Philosophy
We believe:
- Infrastructure should be shared - Everyone needs these tools
- Implementation should be unique - Your business logic is yours
- Community makes better software - More eyes, fewer bugs
- Openness builds trust - See how we work before hiring us
The Bottom Line
Open source isn't charity. It's strategy.
By giving away our tools, we:
- Build trust faster
- Recruit better talent
- Improve code quality
- Set industry standards
- Accelerate innovation
The code is free. The expertise to implement it isn't.
That's a business model that works.
Want to contribute? Check out our repos on GitHub. Want to implement these tools in production? That's where we come in. Let's talk about how open source can accelerate your AI development.
