promptfoo/promptfoo

Features

• LLM Evaluation - Provides a comprehensive framework for testing, benchmarking, and red-teaming language models across multiple providers.
• Prompt Engineering Toolkits - Offers a toolkit for iteratively refining, comparing, and optimizing prompt templates and model configurations.
• Automated Prompt Testing - Evaluation & Testing triggers systematic quality and performance tests for prompts automatically whenever code changes are pushed to a repository.
• Adversarial Red Teaming Toolkits - Automates the detection of jailbreaks, prompt injections, and safety violations by running adversarial test cases against language models.
• Automated Assertion Validators - Validates language model outputs against deterministic rules, semantic similarity metrics, and custom scripts to verify quality and safety.
• Adversarial Security Research - Automates security vulnerability scanning by generating and chaining malicious inputs to probe for jailbreaks and prompt injections.
• LLM Provider Integrations - Connects to various commercial and open-source model APIs to run comparative benchmarks and evaluations within a unified environment.
• RAG Evaluation Frameworks - Assesses the factual accuracy and relevance of retrieval-augmented generation pipelines by comparing responses against source data.
• Stateful Agent Orchestration - Manages multi-turn conversation histories and tool-use trajectories to simulate complex agentic workflows during automated testing.
• Automated Security Scanners - Automates adversarial testing and vulnerability detection for language models and agentic workflows.
• Agent Evaluation Tools - Simulates multi-turn interactions and tool usage to verify agentic task execution.
• Agent Framework Integrations - Connects with orchestration libraries to test, trace, and evaluate multi-step workflows and complex agentic applications.
• AI Model Integrations - Links to a wide range of hosted, local, and custom AI providers through a unified interface for consistent testing.
• Automated Prompt Optimization - Iteratively refines prompts based on performance metrics to identify the most effective versions.
• Validation Frameworks - Assesses retrieval-augmented generation accuracy by comparing model responses against source data.
• Hardware-Agnostic Inference Layers - Standardizes communication across diverse model APIs and local scripts to enable unified testing and comparative benchmarking.
• AI Agent Benchmarks - Simulates complex multi-turn interactions and tool usage to verify agentic workflow reliability.
• Prompt Injection Testing - Generates external content with hidden instructions to evaluate agent manipulation risks.
• Adversarial Test Automation - Executes modular test suites that generate malicious payloads to identify security and compliance risks in language models.
• Evaluation Templates - Structures test cases, prompt templates, and evaluation criteria into portable files for consistent execution across environments.
• Model Observability Suites - Monitoring & Observability calculates perplexity scores for model outputs to quantify prediction certainty and identify potential hallucination risks based on configurable thresholds.
• Test Case Definitions - Provides structured test case definitions to validate model outputs against expected outcomes and assertions.
• Agent Tool Integrations - Provides mechanisms for connecting autonomous agents to external software tools and APIs to extend their functional capabilities during evaluations.
• Model Comparison Interfaces - Provides side-by-side comparison of model versions and prompt templates to identify optimal configurations.
• Provider Configurations - Stores and references API keys and base settings in a central environment while maintaining the ability to override parameters locally.
• AI and LLM Testing - Framework for testing and red-teaming LLM applications.
• AI Frameworks and SDKs - Tool for testing, evaluating, and red-teaming LLM prompts.
• AI Observability and Evaluation - Tool for testing, red-teaming, and evaluating prompts and agents.
• AI Red Teaming - CLI and framework for testing and red-teaming LLM prompts.
• AI Security - Framework for red teaming and evaluating LLM prompts.
• AI Security Frameworks - Framework for red-teaming and evaluating LLM security and performance.
• Evaluation and Observability - Test-driven framework for regression testing and model comparison.
• Large Language Models - Testing and red-teaming for prompts and agents.
• Model Evaluation and Benchmarking - Red teaming and evaluation framework for LLM security testing.
• Moderation APIs - Framework for red teaming and testing LLM safety against attacks.
• Prompt Engineering Tools - Test, evaluate, and compare LLM outputs.
• Prompt Testing and Security - CLI for testing, evaluating, and red-teaming LLM prompts.
• Security & Privacy - Local tool for systematic testing of prompts, performance, and security.
• Continuous Integration Quality Gates - Integrates automated quality gates into CI pipelines to enforce performance standards and prevent regressions.
• Content Guardrails - Validates the effectiveness of safety filters and moderation layers by simulating adversarial inputs.
• Performance Visualization - Provides interactive side-by-side visualization of model outputs to facilitate comparison and manual rating.
• Test Utilities & Assertions - Provides objective and subjective criteria for validating model outputs via assertions and rubrics.
• Test Report Aggregators - Consolidates performance metrics and security findings into structured reports for visualization and integration with development pipelines.
• Automated Test Runners - Orchestrates automated evaluation workflows within CI pipelines to track regressions and report results.
• Tool Definition Adapters - Standardizes tool definitions across multiple model providers to ensure consistent behavior and reliable execution during automated evaluation cycles.
• Conversational Evaluation Suites - Models multi-turn conversation histories to evaluate stateful dialogue flows.
• Evaluation Report Aggregators - Consolidates individual test session results into comprehensive performance reports for centralized tracking.
• External Knowledge Integrators - Connects agents to external databases and APIs to inject domain-specific information for retrieval-augmented generation assessment.
• Data-Driven Testing - Imports and exports test cases and evaluation results using external spreadsheet or document management systems for collaborative data handling.
• AI Governance Policies - Creates bespoke testing plugins to enforce organization-specific behavioral standards and AI governance policies.
• Adversarial Input Transformers - Transforms test inputs using techniques like obfuscation to bypass content filters and security controls.
• Model Vulnerability Scanners - Provides automated scanning of model files to detect security risks and architectural vulnerabilities before deployment.
• Extensible Plugin Architectures - Supports modular extension by allowing developers to inject custom logic for providers, grading rubrics, and attack strategies.
• Language Model Metrics - Uses model-based grading and embedding comparisons to assess factual accuracy and faithfulness.
• Evaluation Grading Configurations - Monitoring & Observability customizes the grading model, scoring weights, and evaluation prompts to tailor accuracy assessments to specific domain requirements.
• Test Report Servers - Distributes test findings and performance metrics to team members through cloud-based platforms or self-hosted infrastructure for collaborative review.
• Test Assertion Extensions - Allows writing custom providers, assertions, and test generators using Python to integrate with external frameworks and libraries.
• Test Case Generators - Automatically generates diverse test cases and personas to expand evaluation coverage.
• Custom Model Logic Interfaces - Allows defining bespoke provider behavior using scripts or local files to test proprietary models or unique workflows.
• Prompt Management Systems - Connects to external version control and observability platforms to monitor, track, and optimize prompt performance.
• Token Bias Adjustments - Forces model responses to adhere to specific choices by applying logit bias to ensure structured and predictable outputs.
• Response Caching - Stores model call results locally to reduce latency and costs during repeated test executions.
• Managed Infrastructure Support - Provides a fully-managed service for hosting evaluation environments to eliminate infrastructure maintenance.
• Self-Hosted Infrastructure - Supports self-hosted deployment within private networks to ensure data sovereignty and control.
• Composite Metric Calculators - Monitoring & Observability calculates composite scores from individual assertion results using mathematical expressions or scripts to generate custom performance indicators.

Star history