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

Engineering at Meta
Engineering at Meta
Microsoft Azure Blog
Microsoft Azure Blog
I
InfoQ
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
人人都是产品经理
人人都是产品经理
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
T
Tailwind CSS Blog
MongoDB | Blog
MongoDB | Blog
Google DeepMind News
Google DeepMind News
WordPress大学
WordPress大学
量子位
美团技术团队
大猫的无限游戏
大猫的无限游戏
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
Last Week in AI
Last Week in AI
博客园 - 司徒正美
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
小众软件
小众软件
C
Check Point Blog
博客园 - 三生石上(FineUI控件)
N
Netflix TechBlog - Medium
Recent Announcements
Recent Announcements
有赞技术团队
有赞技术团队
月光博客
月光博客

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 - 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). mempalace/agent at agent · skorotkiewicz/mempalace
cwmail — terminal email with AI-drafted replies
coolwulf · 2026-06-18 · via Hacker News - Newest: "LLM"

A terminal email client with proper HTML rendering, inline image support, multi-account IMAP, and AI-drafted replies powered by DeepSeek V4 Pro.

Written in Go on Bubbletea v2. Reads real mail. Renders HTML newsletters, receipts, calendar invites, and event posters without dropping out to a browser. Drafts replies you can edit in place.

IMAP

multi-account
Gmail, iCloud, Outlook, custom

IDLE

push notifications
not polling

AI

DeepSeek V4 Pro
drafts your reply

What it is

cwmail is a full-screen Bubbletea TUI for managing IMAP mailboxes. It handles the actual hard parts of email — MIME multipart, HTML rendering, charset detection, inline images, attachments, threading, drafts — not just the obvious ones.

You can manage as many IMAP accounts as you want side-by-side in the sidebar. Folder navigation is fast (Vim-style or arrows), full-text search hits the local SQLite mirror so it returns instantly, and IDLE push keeps your inbox current without polling. Desktop notifications fire when new mail arrives.

The AI-drafted-reply feature is a one-key action: hit R on any open message and the model produces a contextual reply draft in the composer. You can edit, send, save as draft, or discard. Same DeepSeek profile cwcode uses, so one API key powers both apps.

It is not a service. There is no cwmail.io. Your IMAP credentials sit in ~/.config/cwmail/config.json; everything else lives in ~/.local/share/cwmail/. Offline-capable for everything except actually sending and fetching.

Why it’s different

Multi-account IMAP

As many Gmail / iCloud / Outlook / custom-IMAP accounts as you want in the sidebar at once. Each account’s folder tree, unread counter, and IDLE connection is independent. The Spam, Trash, and Drafts folders are merged into a unified view by default.

HTML emails that render

Most terminal mail clients dump the plain-text alternative and call it a day. cwmail’s HTML-to-terminal converter handles tables, horizontal & vertical rules, ordered & unordered lists, block quotes, links, charset translation, and HTML entity decoding. Newsletters, receipts, event invites all read cleanly.

Inline images

Embedded and CID-referenced images render as half-block ANSI in 24-bit color (via pixterm-style rendering). Works on any terminal with true-color support. Useful for event banners, product photos in receipts, and inline charts.

AI-drafted replies

Hit R on any open message. The DeepSeek V4 Pro model drafts a contextual reply (using the same profile cwcode uses). You see the draft in the composer ready to edit. The model knows the prior thread, your typical tone, and whether to be terse or warm based on the conversation. Costs ~$0.01 per draft.

IDLE push, not polling

IMAP IDLE means the server tells the client when new mail arrives. cwmail keeps one IDLE connection per account open and surfaces new mail within a second of receipt. No 5-minute fetch cycles, no batch checking. Desktop notifications fire on arrival.

Undo delete

Delete moves to Trash and shows a brief “undo” toast for 8 seconds. Hit u and the message comes back. Matches matcha.email’s UX for the most common reflex mistake.

Draft auto-save

Any time you’re editing a reply or compose, the buffer is persisted to ~/.local/share/cwmail/drafts/ every few keystrokes. Crash, force-quit, accidentally close the window — the draft is still there next time you launch.

CLI send mode

cwmail send --to alice@example.com --subject “build done” --body “all green” — non-interactive sending, useful for AI agents (cwcode included), CI pipelines, and shell scripts. Reads body from stdin too. Same auth as the TUI.

No SaaS lock-in

Config is JSON. Local mirror is SQLite. Drafts are plain files under ~/.local/share/cwmail/. You can grep your own mail. Credentials live in the OS keyring (macOS Keychain, libsecret on Linux, Credential Manager on Windows).

What it looks like

cwmail inbox view: account sidebar listing Gmail, Boxer, Handle, Inbox 19858, Junk, Sent, Trash, and several folders; main column shows email subjects, senders, and dates in a dense list
Inbox view — multi-account sidebar, dense subject/from/date list, vim-style nav.
cwmail rendering an NVIDIA newsletter HTML email with two embedded event posters shown as ANSI half-block art, event titles, dates, locations, and brief descriptions
HTML newsletter rendered inline — tables, event blocks, and inline images as half-block ANSI.
cwmail reply composer with To, Subject, and Body fields filled in; the body shows an AI-drafted polite reply: 'Thank you for sharing the CUDA sessions at ISC 2026. I'll review the lineup and let you know if I have any questions. Appreciate the update.'
AI-drafted reply — one keypress, contextual draft, then edit and send.

Install

Download a pre-built binary from the Google Drive release folder (current build: v0.9; macOS arm64 / amd64 and Linux amd64). Drop it somewhere on your PATH and make it executable:

curl -L <download-url> -o ~/.local/bin/cwmail
chmod +x ~/.local/bin/cwmail
cwmail --help

Configure your first IMAP account in ~/.config/cwmail/config.json:

{
  "accounts": [
    {
      "name":     "Gmail",
      "imap":     "imap.gmail.com:993",
      "smtp":     "smtp.gmail.com:587",
      "username": "you@gmail.com"
    }
  ],
  "ai": {
    "endpoint":  "https://api.deepseek.com",
    "model":     "deepseek-v4-pro",
    "api_key":   "sk-..."
  }
}

Run it.

cwmail                                # full TUI
cwmail send --to a@b.com --subject hi # non-interactive send
cwmail compose                        # straight to composer

On first launch cwmail prompts for the Gmail app-password (or OAuth flow), stores it in the OS keyring, and starts the IDLE connection.

FAQ

Does it support Gmail OAuth?

App-password is the default and the simplest path. OAuth is wired but requires registering a client; if that matters to you, get in touch.

What about calendar invites?

cwmail parses the ICS attachment and renders the event date, time, location, and summary inline. Hitting Accept sends the reply iCal; declining lets you add a brief note.

Search?

Full-text against the local SQLite mirror. Press / for instant search across subject, body, and sender. Server-side search is also exposed via /i:.

Threading?

RFC 5256 thread sort with In-Reply-To / References fallbacks. Conversations collapse and expand on Enter; navigation walks the tree.

What does the AI reply actually look like?

See the third screenshot above. The model is given the thread context plus a short system prompt that asks for a terse, professional draft. You can configure the tone in the config file. The draft lands in the composer ready to edit — nothing is sent without you pressing Ctrl+S.

Is the source available?

Pre-built binaries on Google Drive. Source currently private — will open once the IMAP edge cases settle. Reach out for early access.

Does it work with cwcode?

Yes — cwcode can call cwmail send as a CLI tool. We use this for “ship a build, email the team” runs at the end of long autonomous loops.