# VizPy

> VizPy is a drop-in DSPy replacement that reduces prompt failure rates by turning errors into executable rules using PromptGrad and ContraPrompt optimizers.

VizPy is a prompt optimization library built as a drop-in replacement for DSPy, designed to reduce prompt failure rates by converting errors into precise, executable correction rules. It offers two core optimizers — PromptGrad and ContraPrompt — that analyze module failures, extract the underlying rules, and inject targeted improvements into instructions in a single API call. VizPy outperforms GEPA on every major benchmark including BBH, HotPotQA, GPQA Diamond, and GDPR-Bench. It is built by VizopsAI, Inc and targets developers building LLM-powered applications who need reliable, automated prompt optimization.

- **PromptGrad Optimizer**: *Runs epoch-based failure analysis, computing textual gradients across batches and accumulating correction rules into a separate local rules section — ideal for generation tasks and continuous metrics.*
- **ContraPrompt Optimizer**: *Mines contrastive pairs (wrong vs. corrected attempts) and extracts the separating rule — best for classification tasks with clear right/wrong labels.*
- **Drop-in DSPy Replacement**: *Pass any `dspy.Module` along with examples and a metric; get back an optimized module with unchanged signature and structure — only the instructions improve.*
- **Plain-English Rule Explanations**: *Every rule added to your prompt is accompanied by a human-readable explanation, making optimization transparent and auditable.*
- **Benchmark-Proven Performance**: *Outperforms GEPA across BBH, HotPotQA, GPQA Diamond, and GDPR-Bench benchmarks.*
- **Simple API**: *Configure with any LLM provider supported by DSPy, instantiate an optimizer, and call `.optimize()` — no credit card required to start.*
- **Scalable Plans**: *Free tier with 10 optimization runs; Pro and Enterprise tiers for higher-volume production use cases.*

## Features
- PromptGrad optimizer for generation tasks
- ContraPrompt optimizer for classification tasks
- Drop-in DSPy replacement
- Epoch-based failure analysis
- Contrastive pair mining
- Plain-English rule explanations
- Single API call optimization
- Benchmark-proven performance over GEPA
- Support for any DSPy-compatible LLM provider

## Integrations
DSPy, OpenAI, Anthropic, Any LLM provider supported by DSPy

## Platforms
WEB, API, DEVELOPER_SDK

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://vizpy.vizops.ai
- Documentation: https://vizpy-docs.vizops.ai/
- EveryDev.ai: https://www.everydev.ai/tools/vizpy