The future built by us.

手把手帮助ai初学者安装正规ai agent。

An open-source skill that lets Claude's Sonnet, Opus, and Haiku share the same memory, so switching models doesn't lose context. Who it's for: People who frequently switch between different Claude models and don't want to repeat background each time. What it solves: Context discontinuity when switching models or starting new conversations. What you get: A set of pure Python, zero-dependency open-source scripts (MIT). v2 triple features: On-demand retrieval (recall.py, TF-IDF search across all historical sessions, works in Chinese and English, only fetches relevant segments); Tiered context (curated long-term memory plus recent sessions kept, older ones auto-archived); Dedup and consolidation (consolidate.py, de-dupes accumulated session summaries, never hard deletes). How to use: Clone the repo, run setup once, then auto-save and load. Works with Cowork, Claude Code, and Chat with Projects. Not included: Doesn't modify your model, doesn't upload data to the internet, everything stays in local files. Open-source skill that gives Claude Sonnet, Opus and Haiku one shared memory store, so switching models doesn't lose your context. v2 adds on-demand recall (recall.py, TF-IDF search across all past sessions in Chinese and English), tiered context (curated long-term + latest sessions loaded, older auto-archived), and consolidation (consolidate.py, de-dupes session dumps, never hard-deletes). Zero dependencies, pure Python, MIT-licensed. Works in Cowork, Claude Code, and Chat with Projects. GitHub: https://github.com/JRcomet/cross-model-shared-memory X: https://x.com/Harry_Atomi/status/2035326204929253718

Corporate websites, personal websites

Hero block design, website. Put back that atomic logo, then adjust the color scheme well. Usable source files, Figma or code.

Who it is for: students and self-learners who study from YouTube / Bilibili / TikTok videos and want reviewable notes instead of rewatching. What problem it solves: watching is slow, and hand-written notes are slower. NoteFlux turns a video link into structured study notes in minutes. What the buyer receives: a personal redemption code worth 3 notes on note.atomi.chat (the hosted NoteFlux web tool). Each note includes a TL;DR summary, a timestamped outline that links back into the video, key terms with definitions, and quiz-style flashcards. Notes are readable on the page and downloadable as Markdown. The code is delivered to your order within 24 hours (usually much faster). How to use it: open note.atomi.chat, enter your code, paste a video link (up to 30 minutes per video), wait 1-3 minutes, read or download your notes. Limits and boundaries: 3 videos per pack, each up to 30 minutes. Subtitle-first pipeline — a failed generation does not consume a note. Personal study use only: no bulk scraping. Source audio/video is deleted immediately after processing; only your notes are kept. Not affiliated with YouTube, Bilibili or TikTok.

10 meticulously crafted GPT-5 Chinese prompt templates. Each template includes role setting, output format constraints, negative examples, and a one-click ready full prompt. Covers four major scenarios: long-form writing, data analysis, programming assistance, and marketing copy. All templates have been iterated over 10+ real tests. Delivery format: Markdown + PDF usage guide.

手把手帮助ai初学者安装正规ai agent。

An open-source skill that lets Claude's Sonnet, Opus, and Haiku share the same memory, so switching models doesn't lose context. Who it's for: People who frequently switch between different Claude models and don't want to repeat background each time. What it solves: Context discontinuity when switching models or starting new conversations. What you get: A set of pure Python, zero-dependency open-source scripts (MIT). v2 triple features: On-demand retrieval (recall.py, TF-IDF search across all historical sessions, works in Chinese and English, only fetches relevant segments); Tiered context (curated long-term memory plus recent sessions kept, older ones auto-archived); Dedup and consolidation (consolidate.py, de-dupes accumulated session summaries, never hard deletes). How to use: Clone the repo, run setup once, then auto-save and load. Works with Cowork, Claude Code, and Chat with Projects. Not included: Doesn't modify your model, doesn't upload data to the internet, everything stays in local files. Open-source skill that gives Claude Sonnet, Opus and Haiku one shared memory store, so switching models doesn't lose your context. v2 adds on-demand recall (recall.py, TF-IDF search across all past sessions in Chinese and English), tiered context (curated long-term + latest sessions loaded, older auto-archived), and consolidation (consolidate.py, de-dupes session dumps, never hard-deletes). Zero dependencies, pure Python, MIT-licensed. Works in Cowork, Claude Code, and Chat with Projects. GitHub: https://github.com/JRcomet/cross-model-shared-memory X: https://x.com/Harry_Atomi/status/2035326204929253718

Corporate websites, personal websites

Hero block design, website. Put back that atomic logo, then adjust the color scheme well. Usable source files, Figma or code.

Who it is for: students and self-learners who study from YouTube / Bilibili / TikTok videos and want reviewable notes instead of rewatching. What problem it solves: watching is slow, and hand-written notes are slower. NoteFlux turns a video link into structured study notes in minutes. What the buyer receives: a personal redemption code worth 3 notes on note.atomi.chat (the hosted NoteFlux web tool). Each note includes a TL;DR summary, a timestamped outline that links back into the video, key terms with definitions, and quiz-style flashcards. Notes are readable on the page and downloadable as Markdown. The code is delivered to your order within 24 hours (usually much faster). How to use it: open note.atomi.chat, enter your code, paste a video link (up to 30 minutes per video), wait 1-3 minutes, read or download your notes. Limits and boundaries: 3 videos per pack, each up to 30 minutes. Subtitle-first pipeline — a failed generation does not consume a note. Personal study use only: no bulk scraping. Source audio/video is deleted immediately after processing; only your notes are kept. Not affiliated with YouTube, Bilibili or TikTok.

10 meticulously crafted GPT-5 Chinese prompt templates. Each template includes role setting, output format constraints, negative examples, and a one-click ready full prompt. Covers four major scenarios: long-form writing, data analysis, programming assistance, and marketing copy. All templates have been iterated over 10+ real tests. Delivery format: Markdown + PDF usage guide.

Hero block design, website. Put back that atomic logo, then adjust the color scheme well. Usable source files, Figma or code.