stanfordnlp/dspy

Health v2 · maintenance only

Adoption not part of the score

37612 stars · 3261 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

DSPy is a Python framework from Stanford NLP for programming language model systems with declarative signatures and modules instead of hand-written prompts. It includes optimizers (like GEPA and MIPRO) that automatically tune prompts and weights of multi-stage LLM pipelines, RAG systems, and agents.

Use cases

• build modular LLM pipelines without hand-writing brittle prompts
• automatically optimize prompts and few-shot examples against a metric
• build RAG pipelines with retrieval and generation stages
• create tool-using agents with ReAct-style loops
• make small cheap models match hand-prompted frontier models
• swap LLM providers without rewriting prompt strings
• fine-tune and prompt-optimize multi-stage programs together
• extract structured data from text with typed signatures

When to choose

• you are building multi-stage LLM programs, RAG pipelines, or agents that need maintainability
• you want automated prompt/weight optimization instead of manual prompt tweaking
• you want provider-agnostic LLM code that works across OpenAI, Anthropic, local models, etc.
• you are an NLP researcher or practitioner exploring new pipelines and tasks

When to avoid

• you need a one-off simple LLM call where a single prompt template suffices
• you want a full application framework with UI, deployment, and chat interfaces built in
• you cannot provide evaluation examples or metrics for the optimizers to work with

Facets

framework · maturity active

agent-framework llm-inference prompt-engineering rag llm-training machine-learning large-language-models artificial-intelligence machine-learning developer-tools python cross-platform prompt-optimization signatures declarative-llm-programs llm-optimizers chain-of-thought react-agents stanford-nlp litellm ai-agents retrieval-augmented-generation natural-language-processing

Member repositories

For agents

markdown · JSON · MCP: product_card(name="stanfordnlp/dspy")