Eventual

Eventual is revolutionizing how developers and data scientists work with complex, large-scale data. At its core is Daft, a groundbreaking, open-source distributed query engine designed specifically for the era of multimodal AI. Backed by a $30M Series A funding round, Eventual aims to provide the foundational technology that makes simple and reliable multimodal data processing a reality, much like SQL did for tabular data decades ago.

The platform addresses a critical bottleneck in AI development: engineers often spend up to 70% of their time wrestling with infrastructure rather than building models. Eventual, with its engine Daft, liberates them from this struggle. It allows for processing petabytes of diverse data types—including images, videos, audio, text, and structured tables—using simple, declarative queries that scale seamlessly from a laptop to a massive cluster.

How to use Eventual

Eventual's core technology, Daft, is designed for developer accessibility. Here's how to get started:

1. Installation: As a Python-native library, Daft can be easily installed via pip: pip install daft.
2. Data Connection: Connect to your data wherever it resides, whether in cloud storage (AWS S3, GCS, Azure Blob Storage), modern data lakehouse formats (Delta Lake, Apache Iceberg, Hudi), or enterprise catalogs (AWS Glue, Unity Catalog).
3. Querying Data: Use Daft's intuitive, Pandas-like API to write queries. The API is declarative, meaning you define what you want to do, and Daft's powerful Rust-based engine optimizes the execution. It supports familiar operations for filtering, joining, and transforming data across different modalities.
4. Applying AI Models: Integrate your Python functions and models from frameworks like PyTorch, NumPy, and HuggingFace directly into your data pipelines using User-Defined Functions (UDFs). Daft's zero-copy mechanism ensures your code runs efficiently next to your data, minimizing overhead.
5. Scaling: Develop your code locally on your laptop. When you need to process larger datasets, the same code can be run on a distributed cluster using a runner like Ray, without any modifications.
6. Managed Service: For enterprises seeking a fully managed, serverless experience, Eventual offers a commercial version. Users can sign up for early access via the official website to offload infrastructure management completely.

Core Features of Eventual

• Unified Multimodal Data Processing: A single, cohesive framework to handle structured tables, unstructured text, images, videos, audio, and tensors.
• Rust-Powered Performance: Built on Rust for exceptional speed, vectorized execution, and efficient memory management. It consistently outperforms traditional systems, using up to 5x less memory.
• Python-Native, No JVM: Avoids the complexity, slow startup times (20x faster), and memory tuning headaches associated with Java-based systems like Spark.
• Seamless Scaling: Write code once and run it anywhere—from a single machine to a large distributed cluster—without changes.
• Universal Data Connectivity: Natively connects to a wide array of data sources and formats, including major cloud providers and data lakehouse technologies.
• Frictionless ML Ecosystem Integration: Works directly with popular ML libraries like PyTorch, Pandas, NumPy, and HuggingFace.
• Intelligent and Reliable: Features smart memory management to prevent out-of-memory errors and sensible defaults to eliminate configuration hassles, ensuring reliable deployments.

Use Cases for Eventual

Eventual and its Daft engine are trusted by leading companies like Amazon, Anthropic, and CloudKitchens for critical data workloads.

• Large-Scale Document Processing: Create text embeddings from massive document collections for applications like semantic search, RAG (Retrieval-Augmented Generation), and text classification.
• Multimodal AI ETL: Build robust ETL (Extract, Transform, Load) pipelines that process and enrich mixed data types, such as annotating images with text generated by a vision model.
• Content Deduplication: Perform fuzzy deduplication at scale on terabytes of text or image data, as demonstrated by Together AI, which achieved a 10x speedup.
• Data Catalog Management: Efficiently manage and process exabyte-scale data catalogs, as proven by Amazon, which saved over 40,000 years of vCPU time annually on a critical job.
• Spark Alternative for ML Platforms: Serve as a more performant and developer-friendly alternative to Apache Spark, unifying infrastructure and improving both query performance and productivity.

Advantages of Eventual

The primary advantage of Eventual is its ability to dramatically boost developer productivity and unlock new AI capabilities. By abstracting away complex distributed systems engineering, it allows teams to focus on innovation. The Rust-based architecture delivers unparalleled performance and efficiency, leading to significant cost savings on cloud infrastructure. Its unified approach to multimodal data breaks down data silos, enabling the creation of more sophisticated and powerful AI systems that were previously impractical to build.

Pricing and Plans

Eventual operates on a freemium model. Its core engine, Daft, is fully open-source and available for free. Developers can install and use it without any cost. For organizations looking for a managed, serverless, and enterprise-grade solution, Eventual offers a commercial product. This managed version is currently available through an early access program, and interested parties can sign up on the Eventual website. Specific pricing details for the managed service have not yet been publicly released.