Tools · SPBoucher
Menu

DSPy

by Stanford NLP and the DSPy community · Framework · also known as dspy-ai, Declarative Self-improving Python

Python framework for programming language-model pipelines with typed signatures and modules, then optimizing prompts and weights automatically against a metric

MIT licenseOpen sourceFreeLocalSelf-hostedSelf-hostable
Visit official site Compare

Link verified Nov 2, 2026

Overview

DSPy is an MIT-licensed Python framework that started at Stanford NLP in 2022 and treats LLM pipelines as programs to compile rather than prompts to hand-tune. A signature declares a task's typed inputs and outputs; modules such as Predict, ChainOfThought and ReAct decide how the signature is executed, including tool use; and optimizers (MIPROv2, GEPA and others) rewrite instructions, select few-shot demonstrations or fine-tune weights to maximize a user-defined metric on example data. Models are addressed through provider-prefixed names, so the same program can target hosted APIs or local runtimes.

The approach suits research groups and teams that need measurable, reproducible pipelines: classification, extraction, retrieval-augmented QA or multi-stage reasoning where a labeled development set exists. Its limits are that optimization consumes many model calls, that results depend on the quality of the metric and data, and that the compiled prompts are less transparent than hand-written ones. The project is at version 3.x and remains actively developed with a lineage of published papers.

Capabilities

  • Typed signatures for task specification
  • Modules: Predict, ChainOfThought, ReAct and custom compositions
  • Optimizers such as MIPROv2 and GEPA for prompts and weights
  • Metric-driven evaluation and compilation
  • Provider-agnostic model addressing

Typical workflows

  1. Compile a classification pipeline against a labeled dev set
  2. Optimize a RAG program's instructions and demonstrations
  3. Compare optimizers on the same metric

Strengths

  • Systematic alternative to manual prompt engineering
  • Research-backed optimizers
  • MIT license and active community

Limitations

  • Optimization is model-call intensive
  • Requires labeled examples and a good metric
  • Compiled prompts are harder to audit

Prerequisites

  • Python 3.10+
  • A model provider or local runtime

Who it is for

  • NLP researchers
  • Engineers building measurable LLM pipelines

Alternatives & comparison

  • Instructor

    Python library for extracting validated structured data from LLM responses using Pydantic models, with automatic retries, streaming partials and support for many providers

    Open sourceOpen source
  • LangChain

    Open-source Python and JavaScript framework whose create_agent harness builds tool-using agents on the LangGraph runtime with swappable model providers and middleware

    Open sourceOpen source
  • promptfoo

    Open-source, local-first CLI and library for evaluating prompts, models, RAG pipelines and agents, with automated red teaming and CI integration

    MixedOpen source
  • Pydantic AI

    Type-safe Python agent framework from the Pydantic team with validated structured outputs, dependency injection, swappable model providers, MCP, durable execution and evals

    Open sourceOpen source
  • Ragas

    Open-source Python library for evaluating LLM and RAG applications with LLM-driven and custom metrics, experiments, dataset management and result tracking

    Open sourceOpen source
Compare DSPy with its alternatives →

Sources & verification

Primary sources consulted for this record. Claims are tied to the fields each source supports; the automated link checker re-verifies every URL on a schedule.

Confidence
High
Reviewed
Nov 1, 2026
Links
Verified · Nov 2, 2026

Spotted an error?

Broken link, outdated license, wrong category, or a tool that no longer exists? Tell us; reports are reviewed by the editor.

Langfuse

Langfuse (part of ClickHouse, Inc. since January 2026) · Prompting, evaluation & observability

In-depth entry

Open-source LLM observability and evaluation platform: tracing, LLM-as-judge and human evals, prompt management, datasets and cost dashboards, self-hostable or as Langfuse Cloud

MITOpen sourceFreemiumSelf-hostedNov 2, 2026

LiteLLM

BerriAI · Prompting, evaluation & observability

In-depth entry

Open-source Python SDK and self-hosted AI gateway that calls 100+ LLM providers in the OpenAI format, with routing, fallbacks, virtual keys, budgets, spend tracking and logging callbacks

MITOpen sourceFreeLocalSelf-hostedNov 2, 2026

OpenRouter

OpenRouter, Inc. · Model providers & APIs

Unified API gateway to hundreds of models from many providers with automatic fallbacks, credit-based pass-through billing, BYOK, free models and data-policy routing

ProprietaryFreemiumUsage-basedBroken

LangSmith

LangChain, Inc. · Prompting, evaluation & observability

LangChain's hosted platform for tracing, evaluating and deploying LLM applications and agents, with prompt tooling, an LLM gateway and cloud, hybrid or self-hosted options

ProprietaryFreemiumPaidSelf-hostedNov 2, 2026

Weights & Biases

CoreWeave · Training & fine-tuning

ML experiment tracking, sweeps, artifacts and model registry plus Weave for LLM tracing and evaluation, now sold by CoreWeave within its Forge platform with a free academic tier

MITFreemiumPaidSelf-hostedBroken

Ragas

Vibrant Labs (formerly Exploding Gradients) · Prompting, evaluation & observability

Open-source Python library for evaluating LLM and RAG applications with LLM-driven and custom metrics, experiments, dataset management and result tracking

Apache-2.0Open sourceFreeLocalBroken