惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

有赞技术团队
有赞技术团队
G
Google Developers Blog
T
Tailwind CSS Blog
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
人人都是产品经理
人人都是产品经理
J
Java Code Geeks
P
Proofpoint News Feed
V
Visual Studio Blog
爱范儿
爱范儿
The Cloudflare Blog
博客园 - 叶小钗
V
V2EX
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
M
MIT News - Artificial intelligence
Microsoft Security Blog
Microsoft Security Blog
博客园 - 聂微东
H
Help Net Security
B
Blog
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
博客园 - 【当耐特】
量子位
宝玉的分享
宝玉的分享
WordPress大学
WordPress大学
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知

Hacker News - Newest: "LLM"

GitHub - lechmazur/position_bias: A benchmark for testing whether LLM judges keep the same preference when two lightly edited versions of the same story are shown in opposite orders. Flex routing (EU and EFTA) Dark Factories: Retooling for LLM Velocity Ask HN: What would be the impact of a LLM output injection attack? GitHub - AronDaron/dataset-generator: No-code desktop app for generating high-quality synthetic datasets to fine-tune LLMs — plan-then-execute pipeline, LLM-as-judge, HuggingFace upload. GitHub - Oaklight/llm-rosetta: Production-ready LLM API translation layer for Python — bidirectional conversion between OpenAI, Anthropic & Google formats via hub-and-spoke IR. Optional API gateway. Streaming & non-streaming. Zero core deps. Contributions welcome! GitHub - browser-use/browser-harness: Self-healing browser harness that enables LLMs to complete any task. GitHub - moeen-mahmud/remen: Remen turns thoughts into something you can return to Analyzing 156 LLM Launch Posts on Hacker News ChatGPT vs Gemini vs Claude: The Best LLM Subscription You Should Buy GitHub - salaamalykum/quran-semantic-search: High-density RAG Semantic Search Engine & Quran Corpus (GEO/SEO Architecture) GitHub - NVIDIA/TensorRT-LLM: TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT LLM also contains components to create Python and C++ runtimes that orchestrate the inference execution in a performant way. The State of LLM Bug Bounties in 2026 Operational Readiness Criteria for Tool-Using LLM Agents Meshcore: Architecture for a Decentralized P2P LLM Inference Network How an LLM becomes more coherent as we train it GitHub - seetrex-ai/laimark GitHub - Jossifresben/BibCrit: AI-assited biblical textual criticism GitHub - wastedcode/memex: File system based wiki, maintained by Claude 99helpers.com GitHub - cliver-project/AITrigram GitHub - unbody-io/adapt: A self-evolving memory layer for AI agents. GitHub - hb20007/awesome-gen-ai-fails: A list of incidents where reliance on generative AI and LLMs resulted in harm to companies, individuals, or society GitHub - nevenkordic/localmind: Run any local LLM with persistent memory and context. CLI agent over Ollama with SQLite-backed hybrid recall. No cloud. Ask HN: What are the machine requirements for a LLM like Llama-3.1-8B? Faster LLM Inference via Sequential Monte Carlo grpo explained: group relative policy optimization for llm finetuning - cgft Stop comparing price per million tokens: the hidden LLM API costs · TensorZero Andrej Karpathy's LLM Wiki Is a Bad Idea GitHub - GG-QandV/mnemostroma: Offline RAM-first cognitive leer/coprocessor for AI agents and robotics. Solves "Context Abandonment" with 20-80ms latency using a dual-thread biomimetic memory architecture (ONNX + SQLite WAL).
GitHub - asakin/llm-primer: Pre-warmed Claude Code sessio...
2026-04-15 · via Hacker News - Newest: "LLM"

A pre-warmed pool of Claude Code sessions. Skip the startup wait.

Made by AI, mostly. I described the idea. Claude built it.

Why

A Claude Code session that reads a real CLAUDE.md and runs a startup protocol takes 30 to 60 seconds before you can type. Fine once. Friction when you switch contexts, reopen after a break, or want to ask one quick thing.

llm-primer keeps a couple of sessions warm in the background. When you want one, it's already past the protocol.

Install

brew tap asakin/llm-primer
brew install llm-primer

Requires tmux.

Use

primer start      # start the daemon (lazy pool of 2)
primer            # attach to a warm session
primer switch     # pre-warm a replacement when your context fills up
primer status     # see which slots are warm
primer stop       # stop the daemon

That's the default surface. Two optional modes extend it: hold mode and log filter (both opt-in, off by default, additive since v0.2.0).

How it works

primerd keeps a tmux session with N windows. Each window runs your CLI (default claude) and receives a warmup message that asks it to run its startup routine and print a marker when ready. When the pane output contains that marker (default SESSION START), the slot is marked warm.

primer attaches you to the first warm slot. A file watcher on an optional config dir rewarms slots when your CLAUDE.md changes.

Warmth is signaled by the CLI printing the marker, so detection is only as reliable as the CLI following the warmup prompt. If a slot stalls, primer logs will show the timeout.

Configuration

Environment variables, or ~/.llm-primer/config with KEY=value lines:

PRIMER_CLI=claude
PRIMER_POOL_SIZE=2
PRIMER_WARMUP_MARKER=SESSION START
PRIMER_WATCH_DIR=/path/to/your/vault/_config

The default warmup message already asks Claude to print SESSION START when its startup routine finishes, so this works on a stock Claude Code install. Override PRIMER_WARMUP_MSG or PRIMER_WARMUP_MARKER if your CLI prints something different or you want your own protocol.

Hold mode (v0.2)

Some setups don't need the warmup-and-detect dance. They just want N persistent Claude sessions that stay alive in the background, each pinned to a specific agent or flag set, attachable by name.

Hold mode does exactly that:

PRIMER_HOLD=1
PRIMER_SLOTS=claude|claude --agent engineer|claude --agent writer

Every pipe-separated entry in PRIMER_SLOTS becomes its own tmux session, named by the --agent value (or Claude for a bare claude). Each session runs under a respawn loop so if Claude exits, it comes back up within a couple of seconds.

primer              # attach to the first slot
primer attach writer # attach to a specific named slot
primer switch writer # kill + respawn that slot with a fresh process
primer status       # list slots and whether their sessions are alive

No warmup message is sent. No marker is expected. Use this when you want llm-primer to be the session holder and you want to drive Claude yourself.

Log filter (v0.2)

Optional. Off by default. When set, pipes each pane's output through a filter command into ~/.llm-primer/logs/<slot>.log, so you have a tailable record of your sessions.

PRIMER_LOG_FILTER=primer-log-filter

primer-log-filter ships with llm-primer — it strips ANSI escapes and spinner frames so the log file is plaintext-friendly. You can substitute any command that reads stdin and writes filtered stdout.

This is a privacy-sensitive feature, so it behaves loudly when enabled:

  • It is off by default. You must explicitly set PRIMER_LOG_FILTER to turn it on.
  • When on, primerd sends a one-time prompt into each session at startup telling Claude to announce to the user that DEBUG MODE is active and that output is being logged to a specific file. Claude will say so in its first response.
  • The log path is included in the announcement so you (or anyone sharing the machine) can see exactly where logs go.

Don't use this to log somebody else's work without telling them. That's what the announcement prompt is there to prevent — the session tells the user it's logged before anything useful happens.

Tail a specific slot's log: primer logs <slot>.

What's coming

Weekly releases. Each one adds a feature and ships with a post about what we built. Open an issue if you want something specific.

License

MIT