이번 νšŒμ°¨μ—λŠ” μΈν¬κ·Έλž˜ν”½ 생성이 μž μ‹œ μ›ν™œν•˜μ§€ μ•Šμ•„ ν…μŠ€νŠΈ λΈŒλ¦¬ν•‘μœΌλ‘œ μ „ν•΄λ“œλ €μš”. λ‹€μŒ μ‹€ν–‰μ—μ„œ μžλ™μœΌλ‘œ λ³΅κ΅¬λ©λ‹ˆλ‹€.
HN TRACTION RADAR
πŸš€ 였늘 HNμ—μ„œ 뜬 μœ μ‚¬ μ„œλΉ„μŠ€ 5건
ν•΄μ»€λ‰΄μŠ€ ν”„λ‘ νŠΈνŽ˜μ΄μ§€μ—μ„œ 골라낸 μ„œλΉ„μŠ€ νŠΈλž™μ…˜ 뢄석
πŸš€ 뢄석 5κ±΄πŸ“‘ 후보 ν’€ 30건
였늘 ν•΄μ»€λ‰΄μŠ€ ν”„λ‘ νŠΈνŽ˜μ΄μ§€μ—μ„œ λ°˜μ‘μ΄ 특히 λœ¨κ±°μ› λ˜ μ œν’ˆΒ·μ„œλΉ„μŠ€λ₯Ό 골라, 각 μ„œλΉ„μŠ€ νŽ˜μ΄μ§€λ₯Ό 직접 읽고 무엇을 ν•˜λŠ” μ„œλΉ„μŠ€μΈμ§€Β·μ™œ λ–΄λŠ”μ§€ λΆ„μ„ν–ˆμ–΄μš”.
였늘의 νŠΈλž™μ…˜ 톱3
πŸ₯‡turbovec
β–² 189 Β· πŸ’¬ 23 Β· ⚑ μ‹œκ°„λ‹Ή 36.6
πŸ₯ˆfx
β–² 43 Β· πŸ’¬ 22 Β· ⚑ μ‹œκ°„λ‹Ή 33.9
πŸ₯‰Cursor Origin
β–² 427 Β· πŸ’¬ 336 Β· ⚑ μ‹œκ°„λ‹Ή 19.5
πŸš€ νŠΈλž™μ…˜ 뢄석
5
πŸ‘‰ μΉ΄λ“œλ₯Ό ν΄λ¦­ν•˜λ©΄ ν•΄λ‹Ή μ„œλΉ„μŠ€ νŽ˜μ΄μ§€λ‘œ μ΄λ™ν•©λ‹ˆλ‹€.
πŸš€ νŠΈλž™μ…˜ 뢄석5
1. turbovec β†—
A Rust vector index with Python bindings built on Google Research's TurboQuant algorithm β€” compresses a 31 GB float32 corpus to 4 GB and searches it faster than FAISS, with no training step. Concrete benchmark claims β€” 3.4Γ— faster than FAISS at 4-bit on both ARM and x86, with 8Γ— memory compression β€” caught the attention of the ML infrastructure crowd looking for FAISS alternatives. β€” πŸ“Š μ„±κ³Ό μ‹ ν˜Έ: 10M-doc corpus: 31 GB float32 β†’ 4 GB compressed; 3.4Γ— avg speedup vs FAISS at 4-bit; 3.5Γ— on ARM; single add latency 6.3–19.7Β΅s (7.6–13.9Γ— faster than FAISS); 362 commits.
β–² 189 Β· πŸ’¬ 23 Β· ⚑ μ‹œκ°„λ‹Ή 36.6
2. fx β†—
A minimal, open-source coding agent CLI written in Zig β€” 6.39 MB binary, model-agnostic, with 10Β΅s cold start, Wasm support, and Unix shell-like ergonomics for research and embedding. Its extreme minimalism (6 MB binary, 10Β΅s cold start, Zig-built) stands out sharply against bloated AI coding tools; 33.9 pts/hr velocity shows rapid early momentum despite only 43 points. β€” πŸ“Š μ„±κ³Ό μ‹ ν˜Έ: 6.39 MiB binary size; 10Β΅s cold start; v0.0.3; 33.9 points/hr velocity.
β–² 43 Β· πŸ’¬ 22 Β· ⚑ μ‹œκ°„λ‹Ή 33.9
3. Cursor Origin β†—
Cursor now hosts Git repositories directly, offering pull requests, code browsing, and two-way GitHub sync β€” all integrated with Cursor's AI agents in one place. Cursor entering GitHub's territory is a major competitive move; with 427 points and 336 comments, the HN community is intensely debating whether this signals a full-stack developer platform war. β€” πŸ“Š μ„±κ³Ό μ‹ ν˜Έ: 427 HN points, 336 comments, 19.5 points/hr velocity. Available to all paid plan users in early beta.
β–² 427 Β· πŸ’¬ 336 Β· ⚑ μ‹œκ°„λ‹Ή 19.5
4. machine0 β†—
Provides persistent CPU and GPU virtual machines (up to 60 vCPUs, 240 GB RAM, H100/H200/MI300X GPUs) billed per-minute, designed for long-running AI agents via CLI or MCP. As a YC S26 company targeting the agent-compute infrastructure gap, machine0 attracted developer interest with its CLI-first design, per-minute billing, and support for NixOS reproducible builds. β€” πŸ“Š μ„±κ³Ό μ‹ ν˜Έ: CPU VMs from $0.013/hr; GPU VMs from $0.836/hr up to 8x H200 at $39.336/hr; 99.99% uptime SLA; 5 regions.
β–² 55 Β· πŸ’¬ 35 Β· ⚑ μ‹œκ°„λ‹Ή 10.1
5. Openleetcode β†—
A local LeetCode runner built in Haskell where test suites and runtime templates live in the open repo; judges solutions locally via a pluggable Docker-based execution backend. Developers frustrated by LeetCode's opaque, proprietary test infrastructure responded positively to a fully transparent, self-hosted alternative with open test manifests and 12 language templates. β€” πŸ“Š μ„±κ³Ό μ‹ ν˜Έ: 60 GitHub stars; 270 commits; 12 supported languages.
β–² 51 Β· πŸ’¬ 14 Β· ⚑ μ‹œκ°„λ‹Ή 7.2
νŠΈλž™μ…˜ μ§€μˆ˜λŠ” ν•΄μ»€λ‰΄μŠ€μ˜ μ‹€μ œ μΆ”μ²œΒ·λŒ“κΈ€ λ°˜μ‘(μ‹œκ°„λ‹Ή ν™”λ ₯)을 κΈ°μ€€μœΌλ‘œ κ³„μ‚°λ©λ‹ˆλ‹€. 뢄석 λ²”μœ„λŠ” ν•΄μ»€λ‰΄μŠ€ ν”„λ‘ νŠΈνŽ˜μ΄μ§€μž…λ‹ˆλ‹€.
somodusWhile I sleep
my AI agent works for meBuild your own AI agent β†’