per on Hacker News

64 titles from 57 authors in the last 30 days, with 61 distinct links and 21 Show HN.

Counts titles, it does not confirm facts. Open the source before concluding anything.

Projects on this term

The score comes from our model.

• Show HN: Vev – Ask questions about a screenshot, get a probability per answer · score 8.0

  Vev is an open-source vision-language model fine-tuned from Qwen3.5 that answers yes/no and multiple-choice questions about screenshots, text, or JSON, returning calibrated probabilities per answer in ~40ms. It implements TypeSafe's Jev API for decision-making, supports images, and ships with benchmarks, a serving CLI, Docker support, and Hugging Face weights.

• Show HN: Zero-shot classifier on scale-to-zero GPUs, one forward pass per call · score 7.8

  Reflex is a GGUF-native Rust and CUDA inference engine optimized for cold starts, compiling all CUDA kernels ahead of time to achieve ~448ms first result on scale-to-zero GPUs. It targets single-tenant, one-shot inference (classification, extraction) where startup latency dominates cost, and includes a serverless sidecar for OpenAI-compatible HTTP.

• Show HN: Semantics to SQL in Rust, ~70µs per query · score 7.5

  0sql is a Rust-based semantic SQL planner that translates business-term query specs into deterministic SQL across 13 dialects, never executing queries itself. It emphasizes governed semantic models, security policies, safe cross-fact blending, and agent integration via MCP, aiming to prevent LLMs from writing raw SQL.

• Show HN: Tallyhook – price Claude Code and Codex logs per client you bill · score 7.5

  Tallyhook is a dependency-free Node.js CLI that parses local session logs from Claude Code and Codex CLI to calculate per-repo, per-team, and per-client AI usage costs, with optional markup for billing. It handles complex log parsing (deduplication, subagents, cache pricing) and offers an MCP server so agents can query their own costs. The project fills a real gap for consultancies and teams needing cost transparency and client billing for AI coding agents.

Recent titles

1. Show HN: Overhear – mention monitoring billed per search, not per month
   2026-11-01 · overhear.dev · HN discussion
2. Show HN: OpenNorme – every law in Italy from 1861 in Git. One law per file
   2026-11-01 · github.com · HN discussion
3. One Durable Object per room, and a rule that stranded a lecture hall
   2026-10-31 · davidmashiah.com · HN discussion
4. Show HN: DeepSeek's flagship got the same answers right and cost 9X per success
   2026-10-31 · github.com · HN discussion
5. Show HN: I build one small site per TikTok trend – here's the latest
   2026-10-31 · tiramisufridgeai.com · HN discussion
6. GDPR: France imposes 27x more in fines per resident than Spain
   2026-10-31 · legiscope.com · HN discussion
7. Does working less explain Europe's GDP per capita gap with America?
   2026-10-30 · philippdubach.com · HN discussion
8. Portugal considers 80 per cent EV purchase subsidies
   2026-10-30 · electrive.com · HN discussion
9. Seychelles has 58,396 IPv4 addresses per 1000 people, the most per capita
   2026-10-30 · en.wikipedia.org · HN discussion
10. Scaling Postgres to 1M durable workflows per second
    2026-10-29 · dbos.dev · HN discussion
11. Is a database per agent (or task) a useful pattern?
    2026-10-29 · github.com · HN discussion
12. How is life at different levels of GDP per capita?
    2026-10-29 · ourworldindata.org · HN discussion
13. Artificial Analysis Index v5, recomputed as dollars per index point
    2026-10-28 · theainewsreport.com · HN discussion
14. 11 Nanoseconds per Word: Over-Optimizing Anagrams
    2026-10-28 · wiredream.com · HN discussion
15. Show HN: Ordpin – 1% per completed order ordering/POS for small venues
    2026-10-28 · ordpin.com · HN discussion
16. Show HN: DocPerRow – one Word document per spreadsheet row, in the browser
    2026-10-27 · docperrow.netlify.app · HN discussion
17. One reviewer per layer: splitting code review by the way code is read
    2026-10-27 · djordjepuzic.com · HN discussion
18. Show HN: Company data from official French and European registers, paid per call
    2026-10-27 · api.sirenic.eu · HN discussion
19. UK to charge universities $1,200 per international student
    2026-10-27 · africa.businessinsider.com · HN discussion
20. How we test LLM serving changes for coding agents, two hours of GPUs per change
    2026-10-26 · upbound.io · HN discussion
21. Poor Man's Claude Pro – All the code you'll ever need for $20 per month
    2026-10-24 · bweller.me · HN discussion
22. Xbox wants you to pay 24 to 40 cents per hour to stream games you bought
    2026-10-22 · arstechnica.com · HN discussion
23. Show HN: Worklicant – Lightweight ATS. Pay per Vacancy. AI Enabled
    2026-10-20 · worklicant.com · HN discussion
24. Show HN: Rouler – Steer Windows traffic per app or IP across multiple interfaces
    2026-10-20 · rouler.io · HN discussion
25. Decoding base64 in Python at 10 gigabytes per second
    2026-10-19 · lemire.me · HN discussion
26. Trillions of operations per second is an incomplete metric for physical AI
    2026-10-19 · newsroom.amd.com · HN discussion
27. Evaluating an AI sales agent on 100 sites/ 15M visitors (+10% rev per visitor)
    2026-10-19 · getjavo.com · HN discussion
28. Show HN: Semantics to SQL in Rust, ~70µs per query
    2026-10-19 · github.com · HN discussion
29. Indra OCR: 1M-parameter OCR model that reads up to 49 pages per second on CPU
    2026-10-18 · github.com · HN discussion
30. ParetoOps – CI gate on cost per successful task for LLM evals
    2026-10-18 · paretoops.dev · HN discussion

All trends