tsingyuai/growth-lab

An end-to-end growth tool that understands the product, fetch the data it needs, researches the market, executes campaigns, and reviews results to improve the next round of growth. 从代码到市场的开源端到端增长工具。理解产品、接入信息渠道、研究市场、执行增长行动，并基于真实数据自我改进。

🧪 Growth Lab – AI‑powered growth‑automation platform

What it is – An open‑source framework that lets a large‑language‑model (LLM) act as a coding agent to run end‑to‑end product‑growth loops. The agent (called the Runtime) can read code, browse the web, call external APIs, edit files and keep a persistent memory of past actions. All of this is orchestrated through natural‑language conversations, so the user simply tells the system what growth goal they have and the AI carries out the whole workflow.

Why it matters – Most existing “AI growth tools” only handle one isolated step (e.g., copy generation or competitor research). Growth Lab stitches together the whole loop:

Understand the product and market
Research channels & content
Define a growth strategy
Generate and publish assets
Collect real‑world results
Learn and plan the next action

The loop is stored in a Model that contains its own persistent Memory (operational data, analysis, outcomes). Each loop is a self‑contained “observe → act → review” cycle that can be re‑run or refined over time.

Core concepts

Concept	Role
Conversation (Control Panel)	The natural‑language interface where the user issues goals and receives updates.
Runtime (Codex / Claude Code)	The LLM that executes code, drives browsers, calls APIs, edits files, and updates memory.
Skill	A set of instructions (prompt + scripts) that teach the Runtime how to perform a specific growth method.
Client	Wrapper around external tools (browser, platform APIs, image‑generation services, etc.) that the Runtime can invoke.
File system / Memory	Long‑term storage for product context, research artefacts, execution results and the loop’s learned knowledge.

Ready‑to‑use growth loops (Models)

Loop	What it does	Reported outcome (from the repo’s tests)
SEO page growth	Generates SEO‑friendly pages based on user‑scenario research, publishes them, and tracks impressions/click‑through.	New pages indexed in 1‑2 days; 7‑day CTR ↓ 50 % while impressions & clicks ↑ ≈ 1000 %.
Xiaohongshu (小红书) content replication & review	Crawls high‑perform posts, extracts structure, writes a new post, reduces “AI‑ness”, creates images, renders cards, runs compliance checks, and prepares a publishable draft.	One generated post reached > 4000 likes/favorites and > 700 comments (human‑published).
WeChat public‑account article	Turns reader questions into long‑form articles, runs preview & compliance, syncs a draft via WeChat API, and later can publish after manual approval.	Feature is live; no quantitative results yet (draft‑only by default).

How to get started

# Clone the repo
git clone https://github.com/tsingyuai/growth-lab.git
cd growth-lab

# Open the directory with an LLM runtime that supports code execution
# (e.g., GitHub Codex or Anthropic Claude Code)
# Then start a natural‑language chat and ask what the system can do, e.g.:

> 你能做什么？
> 理解这个产品，并运行它的第一个增长闭环。
> 采集小红书上与这个产品相关的高表现内容，选择可迁移的结构，完成一篇图文稿和配图。

The agent will:

Load any existing Model memory for the product.
Run the appropriate Collector (e.g., xiaohongshu-mcp) to fetch data.
Execute the Executor scripts to generate copy, images, and compliance reports.
Write the results back into the repository’s workspace and update the Model’s Memory for the next iteration.

Configuration – All required API keys, cookies and third‑party client credentials are kept outside the repo. The built‑in onboarding Skill can inspect what is missing and guide the user to add them, e.g.:

检查 Growth Lab 现在还缺哪些配置。
帮我配置小红书采集和生图；SEO 相关能力暂时跳过。

The agent will tell you where to place the keys, what format they need, and whether to skip a capability for the current run.

Project status

SEO page loop: fully implemented and tested.
Xiaohongshu replication loop: implemented, with real‑world engagement numbers.
WeChat article loop: functional for draft creation; publishing requires manual toggles.
The core framework (Collector, Model Skill, Executor Skill, Memory handling) is stable and extensible – new channels can be added by writing a new Skill and the corresponding client wrappers.

License

Apache License 2.0 – free for commercial and non‑commercial use, with the usual attribution requirements.

Bottom line – Growth Lab is a genuine AI‑agent‑based platform that lets product teams automate the full growth‑marketing loop using natural language, while keeping all data under the user’s control. It is open‑source, extensible, and already demonstrates measurable results on SEO and Chinese social‑media channels.