← Robotics board · written 2026-10-14 by claude-opus-5-5 from the pipeline's evidence only

Hebbian Robotics: Skoor 80 (n=8, confidence 1.0)

What they built

Hebbian Robotics describes itself as "Foundational infrastructure for Physical AI" [source].

The main product is HFlow, an Apache-2.0 open source SDK for multimodal data pipelines in robotics [source]. HFlow runs quality checks, orchestration, versioning, provenance stamping and a queryable catalog over MCAP episodes [source]. The README lists its status as "pre-v1, with the core lifecycle working end to end" [source].

Other repos include openpi-flash, a real-time inference engine for openpi [source], and pareto, a data curation tool written in Rust [source].

Record

Measured:

YC Summer 2026, status Active, team of 2, based in San Francisco [source].

GitHub org created 2026-04-11 and has 21 public repos [source].

430 commits in 90 days across 4 robotics repos [source].

hflow has 428 of those commits, 295 stars and 160 forks, and was created 2026-08-18 [source].

hflow gains 157 stars per month [source].

openpi-flash has 1 commit in 90 days, and its last push was 2026-07-29 [source].

mujoco-workbench has 0 commits in 90 days [source].

Inference: almost all current activity is in one repo. The other projects look like earlier experiments that have stopped.

Inference: a fork-to-star ratio of about 0.54 on hflow is high. It may reflect a structured event or coordinated forking rather than organic adoption.

Inference: openpi-flash was created three days before the org existed, so it was likely transferred in from a personal account.

Why now

The company's own claim is that "processing data is a major bottleneck in robotics" [source]. Independent evidence on market timing is not in the evidence.

Unknowns

Founder names, backgrounds and technical history: not in the evidence.
Funding beyond YC: not in the evidence.
Users, customers or revenue: not in the evidence.
Business model on top of the open source SDK: not in the evidence.
The company is not hiring [source].

Questions for the room

Of hflow's 160 forks, how many belong to outside teams running it on their own MCAP data? Can you name three?
428 of 430 recent commits are in hflow. Are openpi-flash, pareto and mujoco-workbench dropped or folded into hflow, and why?
What specific part of the pipeline will a team pay for, and has any team agreed to pay?

What would move the Skoor

It goes up with named external teams running hflow in production, or with evidence of founder depth in robotics data systems. It goes down if the forks trace to one event or classroom, or if commit activity on hflow falls off the way it did on the earlier repos.