Never Miss an Important Paper in Your Industry Again

Constella scans every paper that comes out and lets you know if it contains any details relevant to your work.

See how it works
Inbox

2 new papers mention your work

Paper alert
Constella Paper Alerts<alerts@constella.app>
to you · This week in your field
9:14 AM

Hi Alex, 2 papers came out this week that touch your research. Here's what's relevant:

1

Retrieval-Augmented Memory Does Not Scale With Context Length

Liang et al.arXiv · 2 days ago · cs.CL
Argues against your thesis, their Section 4 pushes back on the node-count pricing model you're building on. But I noticed some flaws in their eval setup (single-seed, no ablation), so it's worth a look.
2

Effective-Context Degradation Beyond 32k Tokens in Open Models

Okafor & ReyesarXiv · 4 days ago · cs.LG
Reports a 14% effective-context drop past 32k on Mixtral-class models, directly supports the long-context claim in your March lit review.
Both papers added to your canvas, linked to your notes.
Constella scanned 1,284 new papers today and flagged the 2 relevant to you.
Tune what you track · Unsubscribe from paper alerts

From the daily firehose
to the few papers that matter

01

Tell Constella your research area, once.

Set your field and the topics you care about a single time. No feeds to curate, no keywords to babysit. Set it once and forget it.

Your profile
Research areasSaved
Retrieval-augmented memoryLong-context LLMsGraph memory & pricingEval methodology
+Add a topic
02

It reads the thousands of papers published every day.

Constella scans everything that comes out and only alerts you on the handful that are actually relevant to your work. The noise never reaches you.

Today's scan
1,284 papers scanned → 2 relevant to you
Retrieval-augmented memory does not scale with contextFor you
Effective-context degradation beyond 32k tokensFor you
03

Get the summary and key points, written for you.

Not a generic abstract. Constella pulls out the points that touch your work, so you grasp what matters in seconds without opening the PDF.

Summary
Retrieval-Augmented Memory Does Not Scale With Context Length
Liang et al. · arXiv · cs.CL

Key points for you

  • Memory recall degrades past ~16k nodes, regardless of context window.
  • Per-node retrieval cost stays flat, supporting a node-count pricing model.
  • Eval uses a single seed with no ablation, a methodological gap.
Why it matters to you → directly tests the pricing thesis in your March memo.
04

See how it fits what you already know.

Sync your notes apps and sources, and Constella maps each new paper into your knowledge, drawing a green line where it supports you and a red line where it refutes you.

Your knowledge graph

For anyone who can't afford to fall behind

Pick a profile to see the alert they wake up to.

Inbox

2 new papers in your battery research

Paper alert
Constella Paper Alerts<alerts@constella.app>
to you · This week in Materials Science
9:14 AM

Hi Dr. Chen, 2 papers this week intersect your solid-state work:

1

Grain-Boundary Engineering Suppresses Dendrites in Solid-State Electrolytes

Voss et al.Nature Energy · 2 days ago
Supports your approach grain-boundary engineering cut dendrite growth by 40%, matching the mechanism in your latest preprint.
2

Capacity Fade in Sulfide Electrolytes Is Driven by Interfacial Cracking

Ito & BrandtJoule · 4 days ago
Points to interfacial cracking, not oxidation, as the main fade pathway. Worth checking against your cycling data.
Both papers added to your canvas, linked to your notes.
Constella scanned 1,460 new papers today and flagged the 2 relevant to you.
Tune what you track · Unsubscribe from paper alerts

Save hours and the fear of missing out.
Always stay on top, without the stress of always keeping up.

Constella does the reading. You do the thinking.