VizPy

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.