MCAP-based multimodal pipeline
Process multimodal episodes that can combine video, state, actions, timestamps, and metadata from many recording systems, with MCAP as the v1 input and output boundary.
HFlow
HFlow is an open source SDK for building multimodal data-quality, processing, enrichment, and curation pipelines for robotics and physical AI. It helps robotics teams turn recorded episodes into quality-checked, queryable datasets with provenance and catalog-backed curation.
Overview
HFlow is an open source SDK for building multimodal data-quality, processing, enrichment, and curation pipelines for robotics and physical AI. The documentation describes it as production-grade data tooling for turning landed recordings into canonical, quality-checked, queryable datasets.
The project focuses on a common robotics data problem: recordings can include video, state, actions, timestamps, and metadata from multiple systems, but teams need a reliable way to validate quality, run transforms, trace provenance, and curate manifests. HFlow provides that lifecycle around standard MCAP episodes, with Python-based processing steps, catalog-backed evidence, and curation through DuckDB SQL.
Features
Extensible processing steps
Run built-in checks or add your own Python transformations, checks, labels, and enrichments, while keeping existing processing code in your own environment.
Provenance and version tracking
Keep track of how each output was produced with provenance data, schema and tool versions, source URIs when available, and catalog records tied to step versions.
Visible orchestration
Render the workflow as a graph through Airflow DAGs so teams can inspect task status, logs, retries, and reruns during scheduled execution.
Queryable catalog and curation
Store metadata, quality evidence, tags, version stamps, and artifact locations in a Parquet catalog so corpus-wide questions can be answered with DuckDB without opening each recording.
Flexible deployment model
Support both a local single-tenant workspace with Docker Compose and deployment into an existing Airflow 3 environment, keeping the data plane separate from control-plane concerns.
Use Cases
Data quality control for recorded episodes
Validate landed robot recordings by checking for issues such as frozen cameras, drifted streams, missing topics, or duplicate episodes before the data moves downstream.
Porting and formalizing existing processing code
Convert existing scripts into a more auditable pipeline by adding transforms, checks, labels, and enrichments while keeping the underlying Python code and step logic familiar.
Corpus curation and manifest creation
Create curated dataset manifests for training or analysis by querying the catalog, inspecting quality evidence, and writing version-pinned outputs without loading each MCAP file.
Local-to-scheduled pipeline execution
Operate the same pipeline locally during development and then schedule it in an Airflow 3 environment for recurring runs and monitoring.
Dataset audit and troubleshooting
Investigate how a result was produced by tracing provenance, step versions, and artifact locations back through the pipeline graph and catalog records.
Pros and Cons
Pros
- Built specifically for multimodal robotics data, including video, state, actions, timestamps, and metadata.
- Keeps processing code in Python and lets existing code plug in through small adapters instead of a rewrite.
- Records provenance, schema, tool versions, and catalog metadata to help audit and reproduce datasets.
- Supports both local development and scheduled execution through generated Airflow 3 DAGs.
- Lets teams query corpus-level metadata and quality evidence with DuckDB without loading the recordings themselves.
Cons
- The repository is pre-v1, so the implementation status and remaining work still matter when evaluating it for adoption.
- The open-source repository does not include the hosted control plane; the text says that hosted platform interest exists, but it is not implemented here and is not a pre-v1 commitment.
- The source text provided here does not include a full integration list or detailed connector coverage.
FAQ
What is HFlow for?
HFlow is an open source SDK for building multimodal data-quality, processing, enrichment, and curation pipelines for robotics and physical AI. It is aimed at teams that work with recorded robot or sensor data and need a structured way to turn it into curated datasets.
Is HFlow production-ready?
The repository describes HFlow as pre-v1, with the core lifecycle working end to end. It is ready to try locally, and the project points readers to the implementation status and open issues for current details.
What does HFlow process and produce?
HFlow starts from one multimodal episode per supported standard MCAP file. It can run transforms, quality checks, labels, and enrichments, then write canonical MCAP output, provenance, artifacts, and a Parquet catalog for curation.
How can HFlow be run?
The open-source deployment can run as a single-tenant workspace with the included Docker Compose runtime, or as generated Airflow 3 DAGs in an Airflow environment you already operate. The repository also notes a local quickstart and a bundled quickstart.
What integrations does HFlow support?
The source text does not list a fixed integration catalog in the extracted page text. It does mention Python transforms, Airflow 3 DAG generation, DuckDB SQL for curation, the MCAP format, and compatibility with tools such as Foxglove and Rerun for opening MCAP data.
Quick Facts
- Category
- Developer Tool
- Domain
- github.com
- Platform
- Open source GitHub repository
- Primary users
- Robotics and physical AI teams
- Runtime
- Python SDK with Airflow 3 DAG generation
- Data format
- MCAP episodes
Альтернативы HFlow
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repositories and verifies changes with the real compiler, debugger, sanitizers, and tests before showing a diff. It offers a free tier plus paid plans, with editor connectors and zero-retention handling described in the source.
Manta AI
Manta AI is an autonomous web app testing tool for teams that want to map application behavior, catch regressions, and generate tests without writing scripts or maintaining selectors. It works from a URL and supports plain-English test flows, run results with screenshots, and scheduled or deployment-triggered checks.
Mindcase
Mindcase is a web data API and console for extracting structured data from websites and major online platforms. It helps users retrieve clean rows of data through prompt or API access, with prepaid per-record billing.
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads inside Firecracker micro-VMs. It is designed for workflows that need machine-level isolation, private networking between sandboxes, and programmatic control through SDK, CLI, or MCP.
hob
hob is an independent workspace for coding agents that keeps agent sessions, terminals, history, and follow-up work organized around the tools and providers you already use. It is aimed at developers who want local control over routing, history, and workspace structure rather than a bundled model stack.
Ably Chat
Ably Chat is a chat API platform for building custom realtime chat applications. It supports room-based messaging, typing indicators, presence, reactions, and message updates, with usage-based pricing options for different deployment stages.