Stanford Nlp · siet Erste Version als "DSP" im Dezember 2022 veröffentlicht; Umbenennung/Neugestaltung zu DSPy ab August/Oktober 2023. · 9× · tolest 29. Juni 2026
DSPy is an open-source framework developed by the Stanford NLP Group that treats prompts not as brittle strings but as parameterized, compilable Python modules ("programming, not prompting"). Using declarative signatures, modules (e.g., Predict, ChainOfThought, ReAct) and optimizers (e.g., MIPROv2, GEPA, BootstrapFewShot), prompts and sometimes model weights are automatically optimized against a measurable metric. The project is model/provider-agnostic (via LiteLLM) and is actively developed (currently in the 3.x release series), with production use at companies such as Shopify, Dropbox, AWS, and JetBlue.
Features
| Lizenz | MIT-Lizenz |
| Plattform | Python-Bibliothek, Python ≥ 3.10, installierbar via pip |
| Preis | Kostenlos, Open Source (keine Lizenzgebühren) |
| Protokoll-Kompatibilität | OpenAI-kompatible Chat-Completions- und Responses-API via LiteLLM; native Tool-/Function-Calling |
| Release-Datum | Erste Version (DSP) Dezember 2022; DSPy-Ankündigung August/Oktober 2023 |
| Unterstützte Modelle/Provider | Alle LiteLLM-Provider (OpenAI, Anthropic, Google, Azure, Bedrock, HuggingFace, Ollama, lokale Modelle u.v.m.) |