dspy

Automate prompt optimization for multi-step LM pipelines by defining declarative workflows with teleprompters that generate few-shot data and refine prompts automatically. This reduces manual prompt engineering across composed LM calls.

Declarative LM programming defines tasks, signatures, and workflows abstractly rather than writing explicit prompts. Use it for RAG systems and agents to ensure predictable behavior and safe module composition across diverse LM providers.

Yes, you can build self-improving AI agents across diverse LM providers including OpenAI and Anthropic. The declarative architecture supports dependency management and extensible optimization loops for multi-provider agent workflows.

Set up a minimal declarative LM module by defining a task signature and running a basic prediction. The teleprompter-based optimization loop automatically tunes the prompt, providing a quick start for self-improving pipelines.

Automated prompt optimization reduces manual prompt engineering by using teleprompters to generate few-shot data and improve prompts automatically. It provides production-grade modularity with typed outputs for complex AI systems.