BAML Launch Week Day 5

Roadmap to BAML 1.0

We're wrapping up Launch Week with a peek at where BAML is headed in 2025. This isn't just about ticking boxes on a feature list; it's about setting a foundation for how you'll build the next generation of intelligent applications (dare I say… agents 😬).

2024 Recap

As we reflect on the past year, we want to extend our deepest gratitude to the BAML community. None of this would have been possible without you!

Quick Stats:

• GitHub: 8 → 2,000+ stars
• Active developers: 10 → 1,000+
• Discord: 12 → 700+ members
• Core team: 2 → 7 engineers
• Lines of Code: 400k+ lines written

More Awesome Highlights:

• Weekly Releases: We've shipped fast and often, with a Rust core powering everything.
• BAML's SAP + GPT-3.5 beat out GPT-4o + Structured outputs: SAP (schema-aligned parsing) was novel research, and it works even better than expected. 🤯
• VS Code Playground: We've built the first and only VS Code LLM playground.
• Universal LLM Support: Closed-source, open-source—if it's an LLM, BAML speaks its language.
• Language Support: We committed to Tier 1 support for 3 languages (Python, Typescript, and Ruby) and Tier 3 for almost everything else. BAML is being used with over 10 languages, including TypeScript, Python, JavaScript, Ruby, Java, Go, PHP, and Erlang.
• Reliability: When OpenAI went down, BAML customers didn't, because BAML makes fallbacks easy.

2025 Vision: Getting to BAML 1.0

What does a 1.0 release look like? In short, it's not just about new features. It's about making sure that every part of BAML—from the developer experience to the runtime performance—is rock-solid, intuitive, and ready for prime time.

Here's our roadmap:

1. Open & Free
   • BAML will remain open-source, free to use, and have zero mandatory internet dependencies.
2. Ubiquity
   • Tier 1 Support for all major languages and editors.
   • Seamless integration with popular frameworks (Next.js, React, FastAPI, etc.).
   • Simplified Cloud Deployments (AWS, GCP, Azure, Serverless, SST, Next.js).
   • Why it matters: Whether you're deploying a hobby project or an enterprise app, BAML just works.
3. Complete Flexibility
   • Fine-grained control over when and how you invoke any model.
   • Smart Model Selection: Route traffic to the best-fit model automatically.
   • Why it matters: You can optimize performance and cost without having to juggle multiple tools or pipelines.
4. Scale for Teams
   • Native version management. baml-cli update --version $version
   • First-class support for monorepos and multi-BAML projects.
   • A dedicated BAML formatter.
   • Modules, namespaces, and a built-in package manager.
   • Why it matters: As your team grows, you can keep your BAML projects organized, consistent, and easy to maintain.
5. Novel Paradigms
   • Semantic Streaming: Stream responses in a semantically meaningful way, not just token-by-token.
   • Structured Evals & Observability: Built-in logs and metrics to see why a generation turned out a certain way.
   • Automatic Prompt/Model Updates: Type definitions and function calls become data to tune and improve future prompts.
   • Agentic Code: Easily build multi-agent systems with clarity and type safety.
   • Why it matters: BAML goes beyond "just another LLM wrapper" by unlocking patterns you can't replicate elsewhere.
6. Extensibility
   • Transparent architecture so you can understand how BAML works under the hood.
   • Freedom to build your own tooling and custom integrations on top of BAML that others can use.
   • Why it matters: Our biggest wins have come from community-driven ideas and solutions. We want to keep that door open.
7. Open Process
   • A formal BAML RFC (Request for Comments) process, letting developers propose language features, runtime behaviors, or library integrations.
   • Why it matters: We believe BAML should be shaped by the developers who actually use it.

The Next Era of AI Development

Our path to 1.0 is all about stabilizing, clarifying, and pushing boundaries. We want to keep working hand-in-hand with a vibrant community of developers who see LLMs as core building blocks of modern software, not just fancy add-ons.

Thank you for joining us on this journey. We're incredibly excited about what's next and can't wait to see what you'll build!

Ready to dive deeper or contribute? Join our Discord and GitHub communities, and let's make BAML 1.0 a reality together!