04 / 06 Threads 成長分析Threads growth analysis

PulseLion

Threads 行銷成長與個人演算法分析 Agent

A Threads marketing-growth and personal-algorithm analysis agent

不猜 Meta 的演算法，用你自己帳號的數據找出什麼有效。

Don’t guess Meta’s algorithm — use your own account’s data to find what works.

即將開源 · Request accessComing to open source · Request access即將開源 · Request accessOpen soon · Request access Repo 目前為私有Repo currently private
PulseLion icon
模式Mode
展示模式 / 正式模式Demo / live
部署Deploy
Node.js · Docker · Workers + NeonNode.js · Docker · Workers + Neon
模型Models
OpenAI · Anthropic · GoogleOpenAI · Anthropic · Google
技術Built with
  • Vue 3
  • Pinia
  • ECharts
  • Hono
  • Cloudflare Workers
  • PostgreSQL + Drizzle
  • Vercel AI SDK
  • Zod

(01) 為什麼Why

Meta 沒有公開完整的排序權重，網路上的演算法攻略多半是二手推論。PulseLion 把公開的方向當作假設，用你自己帳號的數據去驗證：哪種鉤子、長度、主題、時段、格式在你的受眾身上真正有效。

Meta does not publish its ranking weights, and most “algorithm guides” are second-hand guesses. PulseLion treats the public hints as hypotheses and tests them on your own account: which hooks, lengths, topics, times and formats actually work for your audience.

(02) 06 項items

它做什麼What it does

  1. 01

    成效時間序列Performance time series

    發文後 T+15m 到 T+7d 自動拍快照，計算首小時互動速度。

    Snapshots from T+15m to T+7d after posting, with first-hour engagement velocity.

  2. 02

    特徵擷取Feature tagging

    規則 + LLM 標註開頭鉤子、內容類型、語氣、是否提問等，可人工修正並回饋為範例。

    Rules + LLM tag the hook, content type, tone, whether it asks a question and more — correct them and they become examples.

  3. 03

    個人演算法，附信心區間Your personal algorithm, with intervals

    以近 30 篇中位數為基準，附 Bootstrap 95% 信心區間；少於 8 篇標示「樣本不足」，並區分「相關」與「實驗已驗證」。

    Baselined on the median of your last 30 posts with bootstrap 95% confidence intervals; under 8 posts is flagged “insufficient sample”, and “correlated” is kept apart from “experimentally verified”.

  4. 04

    實驗引擎Experiment engine

    隨機分派實驗組／對照組，達標後檢定顯著性，轉為寫稿規則。

    Random assignment to test and control, a significance test when ready, then a new writing rule.

  5. 05

    寫稿助理Writing assistant

    學習 Top 30 貼文語氣，每個點子產生 3 個鉤子版本，附去 AI 腔檢查與長文拆串。

    Learns the tone of your top 30 posts, writes 3 hook versions per idea, checks for AI-sounding phrasing and splits long posts into threads.

  6. 06

    核准、收件匣、週報Approval, inbox, weekly report

    未經人工核准的草稿永遠不會排程或發布；黃金 60 分鐘收件匣提醒你親自回覆；每週一產出 3 個發現 + 1 個實驗建議，數字全由程式計算。

    Unapproved drafts are never scheduled or published; a golden-60-minutes inbox nudges you to reply yourself; every Monday brings 3 findings + 1 experiment, with every number computed by code.

(03) 怎麼運作How it works

從假設到寫稿規則From hypothesis to writing rule

  1. 01

    同步Sync

    官方 Threads API 拉回貼文與成效快照。

    Pull posts and snapshots via the official Threads API.

  2. 02

    標註Tag

    規則 + LLM 標註特徵，可人工修正。

    Rules + LLM tag features; you can correct them.

  3. 03

    比較Compare

    對照近 30 篇中位數與信心區間。

    Against your 30-post median, with intervals.

  4. 04

    實驗Experiment

    隨機分組、檢定顯著性。

    Randomised groups, significance test.

  5. 05

    核准發布Approve

    只有你核准的草稿會排程。

    Only drafts you approve get scheduled.

(04) README

快速開始Quick start

沒有設定外部服務時以展示模式運作：記憶體資料庫、模擬 Threads API（回應標註 simulated: true）與規則式標籤。

Without external services it runs in demo mode: in-memory database, a simulated Threads API (responses marked simulated: true) and rule-based tagging.

展示模式Demo mode 
pnpm install
cp .env.example .env
pnpm test
PULSELION_DEMO=1 VITE_DEMO=1 pnpm dev
Docker Compose（Node.js + PostgreSQL 16）Docker Compose (Node.js + PostgreSQL 16) 
docker compose up -d --build

(05) 合規承諾Compliance commitments

合規承諾Compliance commitments

  • 不提供自動按讚、自動追蹤或罐頭留言。No auto-likes, auto-follows or canned replies.
  • 只有人工核准的貼文會發布；回覆一律由使用者親自觸發。Only human-approved posts are published; replies are always triggered by you.
  • 只使用官方 Threads API；Token 以 AES-256-GCM 加密並可隨時撤銷。Official Threads API only; tokens are AES-256-GCM encrypted and revocable at any time.
  • 他人回覆預設 90 天後清除。Other people’s replies are purged after 90 days by default.

目前狀態Status

  • 展示模式優先：Threads API 與 PostgreSQL 目前只以 mock 測試驗證過，上線前請用自己的帳號實測。Demo-mode first: the Threads API and PostgreSQL paths are only verified with mock tests so far — test with your own account before going live.
  • LLM 標註已用 claude-haiku-5-5 實測（3 篇 1.7 秒）。LLM tagging was tested with claude-haiku-5-5 (3 posts in 1.7 s).
  • Repo 目前為私有，準備開源中。README 宣告 MIT 授權。Repo currently private — coming to open source. README declares MIT.

想先試用 PulseLion？Want to try PulseLion early?

所有 repo 目前都是私有的，準備開源中。想搶先試用，請透過 GitHub 個人頁聯絡。All repos are private for now and on their way to open source. To try one early, get in touch via the GitHub profile.

即將開源 · Request accessComing to open source · Request access即將開源 · Request accessOpen soon · Request access Repo 目前為私有Repo currently private