Why LitAI?

Agent frameworks like Langchain and CrewAI are great for demos and quick experiments. They come with built-in agents, chains, tools, and flows that help you get something running fast. But that speed comes at a cost. The abstractions are deep, the behavior is unpredictable, and debugging turns into detective work. Even simple things like prompt formatting or tool calling get hidden behind wrappers and internal logic you can’t control. You end up learning the framework instead of building your product. Teams quickly hit the ceiling and either bolt on more tools to regain observability or throw it all out and rebuild their own system from scratch - usually in Python.

LitAI gives you that system from day one. It’s the clean, testable foundation most teams end up writing themselves: model calls, retries, fallbacks, tool use, memory, streaming - all with full transparency and zero glue code. There’s no framework to learn, no YAML, no visual builders, and no magic. You write normal Python programs and use .chat() or .if_() when you want the model to decide something. Everything is observable, everything is debuggable, and everything works the way you expect — so you can move fast without regrets when it's time to scale.

Comparisons

Here's how to think about the main differences.

Feature	LitAI	Most Agent Frameworks
Getting started	One line of Python	Dozens of classes, configs, or YAML
Abstractions	None - use normal Python	Deep, rigid, hidden
Prompt transparency	You control exactly what’s sent	Often modified behind the scenes
Tool use	Just write a function	Requires special classes + patterns
Debuggability	Easy - it's just your code	Hard, opaque layers, magic flows
Testing	Works with pytest out of the box	Requires mocking or special setup
Retry / fallback	Built-in	Often requires extra code or wrappers
Model switching	Per-call, fallback-ready	Possible but complex
Production readiness	Logging, usage, credits, retries	Requires additional tools
What you build	Code you trust	Code you fight