PixelML
Open-source pixelml-av video indexing with Grok answers and Jev relevance plus support checks on long-form video.
by @seanphan
Overview
PixelML (pixelml.com) ships pixelml-av, the open-source library behind Composer and Sentinel agents that index video to text for LLM reasoning. Sean Phan's clip benchmarks Grok plus Jev against Gemini Flash native video on a seventy-five minute file.
Problem: Agentic video Q&A over long files is slow and expensive when every retrieval hop uses a large multimodal model.
Built for: Teams indexing enterprise video libraries with many queries over the same corpus.
First indexed on Jev Directory: 2026-09-20
Creator and team
- Name
- Sean Phan
- Organization
- PixelML
- Handle
- @seanphan
“pixelml-av plus Grok plus Jev: about two tenths of a cent per query and four seconds on eight questions over seventy-five minutes of video.”
How Jev is used
- Role in the product flow
- Relevance filter before Grok answers and support check after
- Primitives
- NoulScore
- State in
- Retrieved candidate evidence with pixelml-av local temporal grouping bound to each query.
- Decision out
- Relevance pass on candidates, Grok answer, then support check with optional frame reinspection when confidence is low.
- Query retrieves candidate evidence from indexed video
- Jev checks relevance with av local temporal context
- Grok generates answer
- Jev support check; low confidence triggers frame inspection
Jev acts as a System One refiner on the search path: fast cheap gates on whether evidence matters and whether Grok's answer is supported, reserving heavier vision calls for low-confidence reinspection. This matches PixelML's stated move away from using the LLM alone to verify every retrieval.
Sourced performance claims
- Eight questions on a seventy-five minute video: about $0.0022 per query and 4.0s for av plus Grok plus Jev vs $0.2253 and 39.9s for Gemini 3.8 Flash native in the clip.Source: Sean Phan X benchmark post
- Described as about 100x cheaper and 10x faster than Gemini Flash native in the same post.Source: Sean Phan X benchmark post
Features and stack
Features
- pixelml-av open-source video indexing library
- Composer and Sentinel agents at pixelml.com
- Grok plus Jev refiner pipeline in public demo
Stack
- pixelml-av
- Grok
- TypeSafe System One
- PixelML agents
Demo
Showcase clip with the same lightbox player as the homepage. Click to play.
pixelml-av relevance with Grok + Jev
Seventy-five minutes of video, eight questions, pennies per query. Jev filters evidence before Grok talks and checks support after so Gemini flash bills do not haunt you.
Links
FAQ
- Is pixelml-av open source?
- Sean Phan's post describes pixelml-av as the open-source library behind PixelML agents. Confirm repo links on pixelml.com.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
- Jev vs LLM classification
When to gate with System One probabilities instead of asking a chat model to label things.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.