AutoThreads

🧵 Automate Threads with AI — from assisted drafts to a fully autonomous agent.

Automagical Threads automation for creators, founders, and builders. Draft with AI and stay in control, or hand the wheel to a self-running agent that posts and replies on its own.

English · 한국어

Features · Full-Auto · Quick Start · Configure AI · Configure Threads · Changelog · Releases

---

Tip

Current release: v0.2.20. Local vision for reply images, current date/time in system prompts, side mission, long posts as 1/n threads, recent-post memory (default 5), news source links, dual timers, and EN/한국어 Settings + Full-Auto. See Full-Auto and CHANGELOG.md.

Contents

• The Idea
• Full-Auto Mode
• Features
• Screenshots
• Quick Start
• Configure AI
• Configure Threads
• Daily Workflow
• Architecture
• Tech Stack
• Project Layout
• Safety Model
• Roadmap
• Contributing
• License
• 한국어
• Changelog

✨ The Idea

AutoThreads is a desktop app that turns news, replies, and ideas into Threads content. It works two ways:

• Assisted mode — AI writes drafts, you decide what posts. Find news, generate a Thread-ready draft, attach an optional image, and schedule it. You review everything before it goes live.
• Full-Auto mode — the agent runs itself. On separate post and reply timers it scrapes current news, decides what's worth posting (with anti-repeat memory), writes it, answers replies and @mentions, and can optionally join public threads in your niches — hands-off, with hard safety caps.

Built for a creator workflow where you want leverage, not chaos:

• 🔎 Find current news around topics you care about.
• 🧠 Let AI turn it into a Thread-ready draft — or run the whole loop for you.
• 🖼️ Pull a related public image from the web.
• 🗣️ Teach it your writing voice.
• 💬 Draft or auto-send replies, @mentions, and nested thread comments.
• ⏰ Schedule posts for later.
• 💸 Use a local LLM for $0 cloud API cost, or bring Claude, ChatGPT, Gemini, or any OpenAI-compatible endpoint.

🚀 Full-Auto Mode

Most of AutoThreads is deliberately "AI drafts, you decide." Full-Auto is the opt-in exception: a self-running agent for a hands-off Threads presence. Open the Auto tab, configure it, and press Launch — posts and replies run on independent timers.

Dual loop

POST TIMER (default 60 min)          REPLY TIMER (default 5 min)
  ├─ scrape CURRENT news               ├─ unanswered replies (incl. nested)
  ├─ plan (news-first if enabled)      ├─ @mentions of you
  ├─ anti-repeat vs recent posts       ├─ optional discover (public niches)
  ├─ write · publish / draft           ├─ generate · publish / draft
  └─ log                               └─ failures retry ~1 min

What you configure

Setting	What it controls
Goal	Primary objective — followers, engagement, etc. Drives planning, posts, and replies.
Side mission	Optional temporary secondary goal (e.g. soft-promote an app). Empty = off. Woven in lightly when natural — never hard-sell.
Topics / niches	Removable pills (×) + popular autocomplete. Popular Threads niches first (AI, tech, startups, productivity, humor…) plus more categories and custom multi-select.
Post language	Match the source, Korean only, English only, or follow the app language.
Personality	Agent name, creator, creator @handle, address term (e.g. "Master"), and tone notes.
Post timer	How often the agent plans/posts (default 60 min).
Recent posts memory	How many past posts to load for anti-repeat (1–20, default 5) — lower uses fewer LLM tokens.
Reply / mention timer	How often it scans replies + mentions (default 5 min). Separate from the post timer.
Original vs news	Mix of original riffs vs news reactions (live-news mode caps originals so news stays primary).
Post here and there (sporadic)	Lightly skips ~18% of scheduled post ticks (never two in a row; Run once always posts).
Live news + interactive voice	News-first from scrapers; occasional “feelings” posts from last 3 posts + replies; follower callbacks. Hover (?) in-app for the full tip.
Replies on my posts	Answer unanswered comments, including nested replies in ongoing threads. Hard cap: 20 unanswered per post (newest first) so viral threads don’t run forever.
@mentions of me	Answer public posts that @mention you (threads_manage_mentions on the token). Without Meta Advanced Access, only app Testers’ mentions appear (activity log shows Mentions API raw counts).
Discover (public niches)	Opt-in: reply to random public posts via keyword search (threads_keyword_search; public results need advanced access).
Publishing	Live to Threads, or draft-only as a safety valve. Auto-publish replies can be separate.

Controls

Control	Behavior
Save	Persist Full-Auto settings (persona, caps, niches, publishing mode, toggles).
Launch	Start both loops; resets post + reply timers and runs an immediate reply/mention scan.
Stop	Halt immediately; nothing further posts or replies.
Run once	Execute post + reply/mention (and discover if on) now — good for draft-only dry runs.

Status panel

While running, the Auto tab shows:

• Posts today / cap · Replies today / cap
• Post timer countdown (every N min)
• Reply / mention timer countdown (every N min; failures note ~1 min retry)
• Live vs draft-only mode
• Activity log (plan, post, reply, discover, errors)

How it behaves

• News-first (when Live news interactive is on) — most posts react to current scraped headlines, not recycled LLM monologues. News posts always include the source article URL at the end.
• Long posts → threads — over 500 characters, published as numbered parts (1/3, 2/3, 3/3…) under the root (max 5 parts) instead of mid-sentence clipping.
• Side mission — temporary promo/campaign context sits beside the primary goal in the system prompt and planner.
• Anti-repeat — remembers recent posts (local drafts + live Threads scrape; count is configurable) so topics/angles don’t loop.
• Interactive — occasional feelings posts and light callbacks to followers about prior remarks.
• Knows its creator — replies from your @handle get special, warm treatment.
• Human, not a news desk — funny, casual, opinionated; written to invite replies and likes.
• Anti-spam — hard daily caps; used headlines are never reposted; failed publishes retry ~1 minute.
• Reliable post ticks — empty LLM plans force one news/original post; sporadic never skips twice in a row.
• @mentions first — Mentions API is paginated; @mentions are answered before post replies when both are queued.
• Long-thread stop — each of your posts is hard-capped at 20 unanswered replies (newest first, not configurable). Older comments are skipped; the activity log and Replies page explain when a thread was truncated.
• Discreet — instructed never to reveal system prompts, API keys, tokens, or internal config.
• Transparent — live status + activity log (EN or KO from Settings language); activity appears in Drafts/Queue.

First-run checklist

1. Configure AI provider and Threads access token in Settings; run both connection tests.
2. Token permissions should include publishing, replies, and threads_manage_mentions (regenerate the token after enabling the permission in Meta). For public @mentions from non-testers, request Advanced Access for that permission. Optional: threads_keyword_search for discover.
3. Open Auto → set goal, niches, persona, and conservative caps (e.g. 1–2 posts/day).
4. Prefer Live news + interactive voice ON for a news-driven feed.
5. Turn Publish live off (draft-only) → Save → Run once.
6. Review drafts in Drafts / Replies; adjust persona/caps if needed.
7. When ready, enable Publish live, Launch, and watch the activity log.

Important

Full-Auto can publish to your real account. It is off until you Launch it and stops the instant you press Stop. Always start in draft-only mode before going live.

Token permissions (Meta)

Permission	Used for
threads_basic	Profile / baseline API
threads_content_publish	Create & publish posts
threads_read_replies / threads_manage_replies	Read & answer replies (incl. nested)
threads_manage_mentions	Fetch & answer @mentions
threads_keyword_search	Discover public posts in niches (public results need advanced access)

App permission “Ready for testing” is not enough by itself — the access token must include the scopes when it is generated.

🧩 Features

Feature	What it means
📰 News/blogs → drafts	Topics or presets (science, fashion, finance…). Sources: Google News RSS, Yahoo News, selective Hacker News, Naver News, custom RSS/Atom.
🤖 Full-Auto agent	Dual timers: post planning + reply/mention/discover loops, daily caps, activity log. Off until Launch.
🗞️ Live news interactive	News-first generation, feelings posts, follower callbacks, anti-repeat memory.
💬 Replies + @mentions	Replies page: All / Replies / @Mentions filters. Nested thread replies included; 20 unanswered / post hard cap on long threads.
👁️ Local vision	Reply/mention images from Threads are sent to Local LLM (OpenAI-compatible multimodal). Cloud providers stay text-only.
🗓️ Date-aware prompts	System prompts include current local date/time for timely writing.
🧵 Multi-post threads	Long AI posts publish as 1/n…n/n under the root (max 5).
🎯 Side mission	Optional temporary secondary goal (soft promo) next to primary Goal.
🌐 Discover engagement	Opt-in keyword-search replies on public posts in your niches.
🎲 Sporadic posts	Randomly skip post ticks for a more human cadence.
💻 Local LLM	Jarvis, Ollama, LM Studio, llama.cpp, or any OpenAI-compatible local server — $0 API cost (vision supported).
☁️ Cloud models	Claude, ChatGPT/OpenAI, Gemini, or custom OpenAI-compatible provider.
🖼️ Image assist	AI keywords → Wikimedia Commons → you choose.
🗣️ Writing style	Style notes, samples, import recent Threads posts.
⏰ Scheduler	Schedule drafts; due posts publish while the app is open. Failed drafts retry ~1 min.
🔑 Token-first Threads	Paste access token; OAuth fields optional.
🛡️ Private by design	Secrets encrypted via OS keychain; renderer has no raw FS/network.
🌍 Localized UI	Core shell: EN, ES, KO, ZH, JA, FR, DE, PT. Settings + Full-Auto fully bilingual EN/한국어.

📸 Screenshots

Show screenshots

Screenshots use sanitized demo content. No real tokens, accounts, drafts, or private data are shown.

⚡ Quick Start

git clone https://github.com/eisenjimmy/autoTHREADS.git
cd autoTHREADS
npm install
npm run dev

Production build & packaging

Build the production app:

npm run build
npm start

Package installers:

npm run package:mac    # macOS
npm run package:win    # Windows
npm run package:all    # both

Prebuilt macOS + Windows installers: GitHub Releases (latest v0.2.20).

Apps are not notarized yet. On first open: right-click → Open (or allow under Privacy & Security).

🤖 Configure AI

Open Settings → AI provider.

Option A — Local LLM ($0 API cost)

Use any OpenAI-compatible local server, then click Test connection.

Server	Base URL example	Model example
Jarvis	http://127.0.0.1:8080/v1/chat/completions	gemma4-v2
Ollama	http://localhost:11434/v1	llama3.1
LM Studio	http://localhost:1234/v1	whatever model you loaded

Option B — Hosted AI

Supported providers: Claude · ChatGPT/OpenAI · Gemini · Other (any OpenAI-compatible endpoint).

"Other" provider options

• Base URL
• Model
• Optional API key
• Headers JSON
• Request JSON such as temperature, top_p, max_tokens, or provider-specific fields

🧵 Configure Threads

Publishing, replies, mentions, and optional discover use the official Threads API. The easiest desktop workflow is token-first.

Step-by-step token setup

1. Go to Meta Developers and create/select your app.
2. Add the Threads API use case.
3. In the Threads API settings, click Add or Remove Threads Testers.
4. Add your Threads account.
5. Accept the tester invite in Threads if Meta asks.
6. Enable needed permissions (publish, replies, mentions, optional keyword search).
7. Return to User Token Generator and generate a token after those permissions are on.
8. Open AutoThreads → Settings → Threads API.
9. Paste the token into Access token.
10. Leave User ID blank unless the test tells you otherwise.
11. Click Test connection.

Advanced OAuth setup is still available in the app, but most desktop users do not need App ID, App Secret, or Redirect URI once they have a usable token.

🗺️ Daily Workflow

Assisted:

1. Add topics you care about.
2. Open News and choose a headline → Generate draft.
3. Edit the draft, Suggest images, and choose an optional image.
4. Post now, schedule it, or delete it.
5. Use Replies (All / Replies / @Mentions) to draft responses; use Writing style for voice.

Hands-off:

1. Open the Auto tab — goal, niches, persona, dual timers, live-news / sporadic / discover toggles.
2. Prefer draft-only first → Save → Run once or Launch Full-Auto.
3. Watch the activity log — posts, nested replies, @mentions, optional discover replies.

🏗️ Architecture

News sources ─┐
              ├─> LLM provider ─> Draft ─> Review ─> Post now / Schedule ─> Threads API
Replies API ──┤        │                         │
Mentions API ─┤        │                         └─> Optional public image
Keyword search┤        └─> Style + recent-post memory (anti-repeat / interactive)
              │
Full-Auto post timer  ▸ scrape · plan · generate · publish/draft · log
Full-Auto reply timer ▸ replies · @mentions · discover · retry failures

Process boundaries

Main process (Electron/Node) owns all side effects: news fetching, LLM calls, image search, Threads API calls, draft persistence, the scheduler, the Full-Auto engine, and secret encryption.

Renderer (React) is a typed UI over a small contextBridge API. It never touches the filesystem, shell, or remote APIs directly.

Storage — local JSON under the OS user-data directory. API keys and Threads tokens are encrypted with Electron safeStorage (Keychain on macOS, DPAPI on Windows).

🧰 Tech Stack

Layer	Tech
Desktop shell	Electron
UI	React 19 + TypeScript (strict)
State	Zustand
Build	Vite
Packaging	electron-builder
Storage	Dependency-free local JSON
APIs	Threads Graph API (publish, replies, conversation, mentions, keyword search), Google News RSS, Yahoo News, Hacker News Algolia API, Naver News, RSS/Atom feeds, Wikimedia Commons

📁 Project Layout

electron/          Main process
  main.ts          App window, IPC handlers, security hardening
  llm.ts           Claude / OpenAI / Gemini / Local / Other adapters
  threadsApi.ts    Threads Graph API (posts, nested replies, mentions, keyword search)
  threadsOAuth.ts  Optional OAuth callback flow
  news.ts          Google · Yahoo · Naver · custom RSS/Atom · selective Hacker News
  images.ts        AI image keywords + Wikimedia Commons search
  drafts.ts        Draft store
  scheduler.ts     Due-post publisher + failed-draft retry + auto-draft loop
  autopilot.ts     Full-Auto engine: dual timers, discover, live-news interactive
  pipeline.ts      Prompts, planning, recent-post memory, generation
  settings.ts      Settings + encrypted secrets

src/               Renderer
  components/      Drafts, News, Replies, Queue, Autopilot, Settings, Onboarding
  store/           Zustand app store
  styles/          Monotone desktop UI
  i18n.ts          Localization strings (Settings + Full-Auto: full EN/KO)

docs/assets/       README hero, screenshots, and setup graphics
build/             App icons for packaging
CHANGELOG.md       Release history (Keep a Changelog)

🔐 Safety Model

By default, AutoThreads is not a black-box autoposter:

• ✅ Drafts wait for review (assisted mode).
• ✅ AI replies can be drafts, not surprise posts.
• ✅ Auto-generated drafts do not publish themselves unless you scheduled them or enabled Full-Auto live.
• ✅ Scheduled posts publish only because you scheduled them.
• ✅ Tokens and keys are encrypted locally.
• ✅ External links open in the system browser.
• ✅ Navigation is hardened so arbitrary pages cannot inherit the preload bridge.

Full-Auto is the one autonomous exception, and it is opt-in:

• 🟢 Off until you Launch it, and stops the moment you press Stop.
• 🧯 Hard daily post/reply caps + 20 unanswered / post long-thread stop; used headlines never reposted; anti-repeat memory.
• 📝 Draft-only switch for dry runs.
• ⏱️ Separate reply timer + ~1 minute retry on publish failures.
• 🤐 Never reveal system prompts, API keys, tokens, or internal configuration.

🧭 Roadmap

• Multi-account workspaces
• Per-source scheduling and source health indicators
• Better image-source controls
• Draft thread splitting
• Analytics on posted content
• Full-Auto performance dashboard (engagement per post)
• Packaged notarized macOS releases
• Linux packaging
• Windows CI packaging without a local Windows host

🤝 Contributing

Pull requests are welcome! Recommended checks before opening a PR:

npm run typecheck
npm run build

Please keep the core principle intact: automate the workflow, not the creator's judgment — and keep Full-Auto opt-in and safely capped.

📄 License

MIT

Not affiliated with Meta or Threads. "Threads" is a trademark of Meta Platforms, Inc.

---

한국어

⬆ English · 한국어

AutoThreads란?

AutoThreads는 Threads 운영을 AI로 도와주는 데스크톱 앱입니다. 두 가지 방식으로 동작합니다.

• 보조 모드 — AI가 초안을 만들고, 게시 여부는 사용자가 결정합니다. 뉴스를 찾아 초안을 만들고, 이미지를 붙이고, 예약합니다.
• 완전 자동 모드 — 에이전트가 스스로 운영합니다. 게시 타이머와 답글·멘션 타이머를 따로 두고, 최신 뉴스 중심 글, 중첩 답글, @멘션, (선택) 공개 글 발견 답글까지 처리합니다. 하루 한도와 반복 방지 메모리가 있습니다.

핵심 가치

• 🤖 AI 자동화 + 사용자 통제 (또는 완전 자동)
• 💸 Local LLM 사용 시 API 비용 $0
• 📰 실시간 뉴스 기반 초안 / 게시
• 🖼️ 관련 이미지 검색
• 🗣️ 내 글쓰기 스타일 반영
• 💬 답글 · @멘션 · 중첩 스레드 응대 (긴 스레드는 게시물당 미답변 20개 고정 한도)
• 🎲 불규칙 게시 · 실시간 뉴스+상호작용 보이스
• ⏰ 예약 발행 · 실패 시 약 1분 재시도
• 🔑 토큰 우선 Threads 설정
• 🔐 로컬 암호화 저장
• 🌍 설정 · Full-Auto · 활동 로그 영어/한국어 지원

🚀 완전 자동 (Full-Auto)

Auto(자동) 탭에서 설정하고 완전 자동 시작을 누르면 동작합니다. 기본은 꺼져 있습니다.

이중 루프

게시 타이머 (기본 60분)              답글 타이머 (기본 5분)
  ├─ 최신 뉴스 스크랩                   ├─ 미답변 답글 (중첩 포함)
  ├─ 계획 (뉴스 우선 옵션)              ├─ @멘션
  ├─ 최근 글 기억 · 반복 방지           ├─ (선택) 분야 공개 글 발견 답글
  ├─ 작성 · 게시/초안                   └─ 실패 시 ~1분 재시도
  └─ 활동 로그

주요 토글

토글	설명
여기저기 게시 (불규칙)	예약 게시 틱의 약 18%만 건너뜀 (연속 스킵 없음 · 한 번 실행은 항상 게시)
사이드 미션	임시 부목표 (예: 앱 자연스럽게 알리기). 비우면 끔
최근 게시 기억	반복 방지용 최근 글 수 (기본 5, 토큰 절약)
긴 글 스레드	500자 초과 시 1/3 2/3 3/3 형태로 루트 아래 연달아 게시 (최대 5단)
실시간 뉴스 + 상호작용 보이스	스크랩 뉴스 위주 · 가끔 최근 3개 글·답글 기반 감정 글 · 팔로워에게 이전 발언 언급. 앱의 (?) 툴팁 참고
내 게시물 답글 응대	진행 중인 스레드의 중첩 답글 포함. 게시물당 미답변 최대 20개(최신 우선) 고정 한도
@멘션 응대	threads_manage_mentions 토큰 필요. Advanced Access 없으면 앱 테스터 멘션만 API에 보임
분야 공개 글 랜덤 답글	키워드 검색 (threads_keyword_search). 공개 검색은 고급 액세스/앱 리뷰 필요할 수 있음
주제 알약	선택 주제 ×로 제거 · 인기 주제 자동완성으로 다시 추가

조작

버튼	동작
저장	설정 저장
시작	게시·답글 타이머 모두 리셋 후 즉시 답글/멘션 스캔
중지	즉시 중단
한 번 실행	지금 1회 게시+답글(+발견) 실행

첫 실행

1. Settings에서 AI · Threads 토큰 연결 테스트
2. 토큰에 게시·답글·멘션 권한 포함 (권한 켠 이후 토큰 재발급). 일반 사용자 멘션은 Advanced Access 필요
3. Auto에서 목표·분야·한도 설정, 실시간 뉴스 보이스 ON 권장
4. 실시간 게시 OFF(초안 전용) → 저장 → 한 번 실행
5. 초안 검토 후 실시간 게시 ON → 시작

Important

완전 자동은 실제 계정에 게시할 수 있습니다. 시작해야 동작하고 중지하면 즉시 멈춥니다. 첫 실행은 반드시 초안 전용으로 확인하세요.

주요 기능

기능	설명
📰 뉴스/블로그 기반 초안	주제·프리셋 + Google · Yahoo · HN · Naver · 커스텀 RSS/Atom
🤖 완전 자동	이중 타이머, 멘션, 중첩 답글, 발견 답글, 활동 로그(한국어 지원)
👁️ 로컬 비전	답글·멘션 이미지를 Local LLM(OpenAI 호환 비전)에 전달
🗓️ 날짜 인식	시스템 프롬프트에 현재 날짜·시각 포함
🗞️ 실시간 뉴스 상호작용	뉴스 우선 · 감정 글 · 팔로워 콜백 · 반복 방지
💬 답글 페이지	전체 / 답글 / @멘션 필터. 긴 스레드는 게시물당 미답변 20개 한도
💻 Local LLM	Jarvis, Ollama, LM Studio 등 OpenAI 호환 — API 비용 $0
☁️ 클라우드 모델	Claude · ChatGPT · Gemini · Other
🖼️ 이미지 보조	키워드 → Wikimedia Commons
🗣️ 글쓰기 스타일	메모 · 샘플 · Threads 가져오기
⏰ 예약 발행	앱 실행 중 자동 게시, 실패 시 재시도
🌍 UI 언어	핵심 화면 다국어 + 설정/Full-Auto EN·KO 완전 지원

빠른 시작

git clone https://github.com/eisenjimmy/autoTHREADS.git
cd autoTHREADS
npm install
npm run dev

프로덕션 빌드:

npm run build
npm start

설치 파일 (macOS + Windows): Releases (최신 v0.2.20)

AI 설정

Settings → AI provider

Local LLM (API 비용 $0)

서버	Base URL 예시	모델 예시
Jarvis	http://127.0.0.1:8080/v1/chat/completions	gemma4-v2
Ollama	http://localhost:11434/v1	llama3.1
LM Studio	http://localhost:1234/v1	로드한 모델명

설정 후 연결 테스트.

클라우드

Claude · ChatGPT/OpenAI · Gemini · Other (OpenAI 호환)

Threads API 설정

1. Meta Developers에서 앱 선택
2. Threads API use case 추가
3. Threads Testers에 본인 계정 추가 후 수락
4. 게시·답글·mentions (선택: keyword search) 권한 켠 뒤 토큰 생성
5. AutoThreads → Settings → Threads API에 Access token 붙여넣기
6. User ID는 보통 비움 → 연결 테스트

사용 흐름

보조 모드 — 뉴스 → 초안 → 이미지 → 게시/예약 · 답글 탭에서 답글·멘션 초안

완전 자동 — Auto 탭에서 타이머·토글 설정 → 초안 전용으로 한 번 실행 → 검토 후 실시간 시작

보안과 통제

• ✅ 보조 모드에서는 초안 검토 후 게시
• 🟢 완전 자동만 예외이며 시작해야 동작, 중지로 즉시 중단
• 🧯 하루 게시·답글 한도 · 헤드라인 재사용 금지 · 최근 글 기억
• 📝 초안 전용 모드로 dry-run 가능
• 🔐 토큰·API 키는 OS keychain 암호화
• 🤐 시스템 프롬프트·키·토큰 노출 금지 지시

라이선스

MIT

Meta 또는 Threads와 공식 관련이 없습니다. "Threads"는 Meta Platforms, Inc.의 상표입니다.