frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: Morph Reflexes – Multi-head classifiers for agent traces

12•bhaktatejas922•1d ago
The most common failures for production agents are behavioral: looping, reasoning leakage, user frustration, and more. Using a frontier model like GPT or Sonnet to judge every turn is too expensive and slow to run at scale.

To solve this, we built Reflexes: semantic signals from agent traces, served fast and cheap over API. Built on custom kernels and a custom inference engine forked from vLLM.

Under the hood, it is a small LLM architected around multi-head inference. Small models need to be trained for specific tasks, but running 50 separate small models on the same input for 50 tasks makes no sense.

How it works: We use a modern LLM with hybrid attention and remove the decode step. We built an inference engine that lets prefill compute be 99% reused from reflex to reflex, similar in spirit to older 2019-era BERT/HYDRA and older multiple-head techniques. we built the inference engine to reuse the KV/cache across inputs and compute across all reflexes. One shared backbone reads the trace once, then many heads classify different signals. Our inference engine reuses the same KV/cache and compute across all reflexes, giving us sub-30ms inference with less than 0.1% overhead for each additional reflex.

We took the same high-level idea and did the hard work to make it work with a modern architecture and attention. On it, we can run inference in under 30ms and serve the full request in under 90ms. If you run 4 reflexes or 100, the extra overhead is less than 2ms.

Why does optimizing this matter?

If you’re even a medium-sized startup, you’re dealing with tens of thousands of agent runs and millions of turns. If you want to track things like user frustration rates over time, frontier LLM-as-judge does not scale.

I built a similar stack at Tesla. When ML engineers needed to sample data across petabytes for signals like `is_camera_obfuscated=true`, along with 200 other things, you need to 1) spin them up quickly 2) run at scale efficiently

What it is not: A dashboard. 99% of dashboards go unused. 100% API first and made for devs who want to use this to trigger their own stuff.

vibetrain a custom reflex in our dashboard, and/or then let it self improve in production: https://www.morphllm.com/dashboard/reflex

Docs: https://docs.morphllm.com/sdk/components/reflexes/index

I’d love feedback from people running agents in prod: what sorts of things do you wish you could track over time across 100% of turns but cant right now?

TLDR: semantic signals from agent traces, super fast, cheap via API

Comments

teitoklien•23h ago
I've been a user of Morph Apply,

Love your products and team ! I hope you guys grow a ton

bhaktatejas922•4h ago
thanks!

Meta Caps Internal AI Token Spending After Costs Approach Billions in 2026

https://mlq.ai/news/meta-caps-internal-ai-token-spending-after-costs-approach-billions-in-2026/
72•typeofhuman•1h ago•58 comments

ZCode – Harness for GLM-5.2

https://zcode.z.ai/en
193•chvid•3h ago•206 comments

Show HN: Searchable directory of 22k+ products from worker-owned co-ops

https://www.workerowned.info/
211•IESAI_ski•4h ago•34 comments

For first time, a cell built from scratch grows and divides

https://www.quantamagazine.org/for-the-first-time-a-cell-built-from-scratch-grows-and-divides-202...
729•defrost•11h ago•249 comments

Building an Open-Source Robot Vacuum – Meet Oomwoo

https://makerspet.com/blog/building-an-open-source-robot-vacuum-meet-oomwoo/
22•devicelimit•56m ago•1 comments

What to learn to be a graphics programmer

https://blog.demofox.org/2026/07/01/what-to-learn-to-be-a-graphics-programmer/
241•atan2•7h ago•123 comments

Opening up 'Zero-Knowledge Proof' technology to promote privacy in age assurance

https://blog.google/innovation-and-ai/technology/safety-security/opening-up-zero-knowledge-proof-...
51•consumer451•3h ago•30 comments

The Underhanded C Contest

https://underhanded-c.org/
33•ccabraldev•3h ago•5 comments

Physical disc production ending in Jan 2028 for new games on PlayStation

https://blog.playstation.com/2026/07/01/physical-disc-production-ending-in-january-2028-for-new-g...
595•Tiberium•13h ago•628 comments

FFmpeg 9.1's new AAC encoder

https://hydrogenaudio.org/index.php/topic,129691.0.html
282•ledoge•11h ago•94 comments

Chip Off the Old Block

https://www.astralcodexten.com/p/chip-off-the-old-block
40•paulpauper•4h ago•5 comments

Qualcomm Linux 2.0

https://www.qualcomm.com/developer/blog/2026/06/qualcomm-linux-2-now-available
50•gilgamesh3•4h ago•12 comments

Global review confirms mRNA vaccines are safe, effective and full of promise 

https://news.ubc.ca/2026/06/mrna-vaccines-are-safe-effective-and-full-of-promise/
65•coloneltcb•1h ago•33 comments

The <Usermedia> HTML Element

https://developer.chrome.com/blog/usermedia-html-element
23•twapi•1h ago•14 comments

Proliferate (YC S25) Is Hiring

https://www.ycombinator.com/companies/proliferate/jobs/mMHvKR9-founding-product-engineer
1•pablo24602•4h ago

Box3D, an open source 3D physics engine

https://box2d.org/posts/2026/06/announcing-box3d/
412•makepanic•13h ago•92 comments

Internal Combustion Engine (2021)

https://ciechanow.ski/internal-combustion-engine/
284•StefanBatory•12h ago•73 comments

Ask HN: Who is hiring? (July 2026)

162•whoishiring•10h ago•175 comments

Monetization Gateway: Charge for any resource behind Cloudflare via x402

https://blog.cloudflare.com/monetization-gateway/
249•soheilpro•11h ago•165 comments

How do wombats poop cubes? Scientists get to the bottom of the mystery

https://www.science.org/content/article/how-do-wombats-poop-cubes-scientists-get-bottom-mystery
35•bushwart•1d ago•7 comments

I Left Harry's All-Night Hamburgers

https://escapepod.org/2013/09/14/ep413-why-i-left-harrys-all-night-hamburgers/
70•rbanffy•4h ago•8 comments

Healthy but sedentary people show early decline in cellular energy production

https://news.cuanschutz.edu/news-stories/healthy-but-sedentary-individuals-show-early-decline-in-...
56•littlexsparkee•2h ago•46 comments

Visual Basic on the PC with Windows 3.1

https://stonetools.ghost.io/visualbasic-win31/
7•TMWNN•3d ago•2 comments

Flavor Graveyard

https://www.benjerry.com/flavors/flavor-graveyard
24•NaOH•3d ago•12 comments

The Apple Disk II Controller Card

https://www.bigmessowires.com/2021/11/12/the-amazing-disk-ii-controller-card/
42•stmw•2d ago•10 comments

Launch HN: Parsewise (YC P25) – Reason Across Documents with an API

45•gergelycsegzi•11h ago•45 comments

How We Made IPFS Content Publishing 10x Faster

https://probelab.io/blog/optimistic-provide/
146•dennis-tra•10h ago•48 comments

Fable 5 Is Back

https://twitter.com/claudeai/status/2072402636813607381
318•mfiguiere•6h ago•296 comments

Weave Robotics launches Isaac 1, a $7,999 home robot with Fall 2026 deliveries

https://www.weaverobotics.com/isaac-1
73•ryanmerket•7h ago•123 comments

Ask HN: Who wants to be hired? (July 2026)

107•whoishiring•10h ago•252 comments