← Robotics board · written 2026-10-17 by claude-opus-5-5 from the pipeline's evidence only

General Instinct: robotics, Skoor 80 (n=6, confidence 1.0)

What they built

General Instinct built InstinctFlash, an open-source serving runtime for robotics models, licensed AGPL v3 [source]. It serves eight robotics model families through one Python and WebSocket runtime on Jetson Thor, RTX 5090 and RTX 4090 [source]. YC lists the company as "Deployment infrastructure for Physical AI" and "Making robot post-training and deployment scalable" [source]. A second repo, InstinctRazor, is an LLM compression framework with 0 commits in 90 days and a last push on 2026-06-27 [source].

Record

Measured:
  • InstinctFlash had 281 commits in 90 days, with a last push on 2026-09-27 [source].
  • InstinctFlash has 200 stars and 6 forks, and InstinctRazor has 69 stars and 7 forks [source] [source].
  • The GitHub org was created 2026-05-28, has 3 public repos and is not verified [source].
  • The company is in YC Spring 2026 with a team of 2 in San Francisco [source].
  • There is one Greenhouse post titled "Career Opportunities" with location "Any" [source]. YC lists the company as not hiring [source].

    • Self-reported benchmarks:

  • Same-schedule speedups on Jetson Thor range from 1.19× (GR00T N1.7, bit-exact) to 7.88× (pi05, FP8) [source].
  • The 33.78× headline includes reducing LingBot-VA from 25V/50A to 2V/4A [source].
  • Several rows were measured with engines that have since been replaced, and the current source has not repeated those measurements [source].
  • The "no observed loss in task performance" claim rests on the team's own real-robot tests, and the excerpt shows no task data [source].

    • Inference: the Greenhouse post reads as a general application form, so the hiring signal (weight 20) may overstate hiring. Inference: three of the six signals come from the YC page, so the Skoor depends heavily on one source.

    Why now

    The team released full source on 2026-09-15, added RTX 4090 on 09-16 and RTX 5090 on 09-17, and shipped Jetson Thor engines on 09-26 [source]. Inference: the bet is that robot models such as pi05 and GR00T need dedicated serving on edge hardware. Market size and demand data: not in the evidence.

    Unknowns

  • Founder names and backgrounds: not in the evidence.
  • Funding: not in the evidence.
  • Customers and revenue: not in the evidence.
  • Production users of InstinctFlash: not in the evidence.
  • How AGPL licensing converts to revenue: not in the evidence.

    • Questions for the room

  • The 33.78× headline combines FP8 with fewer sampling steps. What is the speedup at the original schedule on current code, and what task success rates support "no observed loss"?
  • Of the 200 stars and 6 forks, who runs InstinctFlash on a real robot outside your team?
  • InstinctRazor stopped in June, and the job post names no role. Are both founders now on serving only, and what is the first hire?

    • What would move the Skoor

    Up: benchmarks re-run on current source, and named external deployments. Down: speedups that disappear at the original schedule, or no users beyond the two founders.