DSPy

Verified 19 Aug 2026

Stanford’s framework for programming — not prompting — LMs.

DSPy is an open-source framework for compiling LM programs: signatures, modules, and optimizers instead of ad-hoc prompts. It is used as an orchestration/intent layer when the factory must change prompts without rewriting folklore.

Orchestration Intent

Model independence: Multi-model
Deployment: Self-hosted
Open source: Yes
License: MIT
Pricing model: Free
Added / updated: 19 Aug 2026 / 19 Aug 2026

dspy.ai · Docs · Source

Editorial take

DSPy is how you stop treating prompts as YAML folklore. It will not open a PR. Use it when the factory’s quality work is “make this program better against a metric,” not when you needed a worker VM.

Strengths

Programmatic prompts with optimizers and a public MIT repo.
Model-agnostic.
Forces a metric, which most prompt files never do.

Limitations

Research-shaped learning curve.
Not an execution sandbox.
Optimizers can overfit a tiny eval you did not hold out.

Fields last checked against primary sources on 19 Aug 2026. Pricing and plan names rot; follow the vendor URL.