syftr
Open source framework that rapidly searches AI agentic workflow configurations to determine the optimal structure, components, and parameters, tailored to your specific data and use case.
In industry-standard RAG benchmarks syftr identifies workflows that cut costs by up to 13x with only marginal accuracy trade-offs—delivering near-optimal performance at a fraction of the price.
Discover optimal agent pipeline patterns, components, and their parameters
Rapidly experiment with multiple hyperparameters to identify Pareto-efficient flows that improve core performance metrics without degrading competing objectives.
Run computations efficiently with minimized costs
Only spend compute resources evaluating top performing agentic workflows with early stopping.
Painlessly evaluate and implement the newest techniques and models
Generate production-ready pipelines of the most performant combination of new and existing technologies without endless testing and benchmarking cycles.
Find optimal balances between accuracy, latency, and cost using Pareto evaluation, ensuring no objective is compromised.
Reduce compute costs by 60-80% through Bayesian optimization methods that eliminate suboptimal flows.
Deliver component-agnostic evaluations and production-ready pipeline code for the most nacent models and workflow techniques, integrating both built-in components and open source contributions.
Whether you are looking for overall improvements, maximum accuracy or minimized costs, syftr can identify new Pareto optimal flows or confirm the quality of your existing baseline.
What’s included in the search space:
Proprietary and open source LLMs
Embedding models
Prompt strategies
Retrievers
Text splitters
Benchmark datasets
Dive into the readme and repo to kick off your first search