saas-growth-agents README

The project's README from github.com/Pluviobyte/saas-growth-agents. As one markdown file: README.md.

saas-growth-agents

saas-growth-agents

Hand your SaaS growth chores to an agent that runs them every day.

Spot new search terms, find who ranks for them, check competitor backlinks, submit to directories, and file good posts into a knowledge base.
These jobs repeat daily and follow fixed steps, which makes them perfect SOPs for an agent.
Create a bot, send it one sentence, and it reads this repo, asks you a few questions, and sets itself up.

Stars License Python Data
Grok Bot Muse Dot Hermes OpenClaw Claude Code Codex

Quick start  ·  The bots  ·  One-line setup  ·  AGENTS.md  ·  Grok Bot free for 14 days  ·  English · 简体中文


Data flow: sources go through AIsa into three script pipelines (new-term radar, site ops, social analysis). Scripts only do deterministic work such as collecting, deduping and logging. The agent analyzes, verifies and reports, then pushes to you to decide. State lives in files and ledgers that are read before every run. Spending, sending and deleting all need your OK.

These are the SOPs and scripts the author runs for their own SaaS tools aimed at overseas markets. Any agent that can run terminal commands and keep long-lived instructions can use them.

Coming from douyin-comments? This repo is the old douyin-comments, now merged into the full growth system. The Douyin comments skill is unchanged and lives in the social analysis pack at bots/social-kb/douyin-comments/. See how to reinstall.

The whole system runs on three rules:

  1. Scripts do the deterministic work: collecting, deduping, scoring and logging. Results are reproducible and replayable, never a model's "feeling".
  2. The agent judges: it reads what the scripts produce, then analyzes, extracts, verifies and reports. If data is missing it says so instead of making things up.
  3. You decide: the agent only pushes conclusions. Spending money, sending anything out and deleting all need your OK. Every correction you make becomes a new rule in its skills.

The bots

Bot What you do What it does
New-term radar Nothing Three reports a day on newly searchable terms: new AI products, models, memes, templates. Sources include Google Trends, competitor sitemaps, new directory listings, Product Hunt, GitHub Trending, Reddit and TikTok
Site ops Give it keywords and your product Finds every site competing for a keyword, checks their DR and backlinks, tracks your own site's metrics daily, submits to free directories by the rules, and logs everything
Social analysis pack Drop a link when you see something good Fetches the original (X, Douyin, RedNote, YouTube, WeChat, any web page) → summarizes → extracts "what could improve your SOPs" against your own workflow → files it by topic. For Douyin it can also pull the full comment section

Plus an unrelated extra, Trade radar: watches price moves and breaking events and pushes an analysis. It analyzes only and never places orders.

Which agent

Grok Bot, Muse, Dot, Hermes and OpenClaw all work, and so do Claude Code and Codex. AGENTS.md has the generic steps; each product just has its own name for "routines" (scheduled tasks) and "vault" (secret storage).

The author recommends Grok Bot. It comes with a cloud computer, a vault, skills and routines. These SOPs were first built on Grok Bot, and once installed they run around the clock on routines.

No Grok Bot yet? Sign up with the author's invite link to get 14 days free: https://grok.com/refer/bot?code=NASHTX8NMYH3 (This is a referral link and the author is rewarded for sign-ups. Grok's official page has the current terms.)

Quick start

  1. Create a new bot in your agent.

  2. Send it the matching sentence from bootstrap.md, for example:

    Read AGENTS.md in https://github.com/Pluviobyte/saas-growth-agents and set yourself up as the newterm (new-term radar) bot. Ask me one question at a time.
    
  3. It asks you to put your AIsa key in its vault, does one run by hand to show you, and only turns it into a routine once you're happy.

The whole process is do it once → tweak until it's right → save as a skill → schedule as a routine.

The job descriptions and SOP docs inside the repo are written in Chinese. Agents read them fine, and the bot will talk to you in your language.

Where the data comes from

All external data goes through AIsa: one key for thousands of APIs, pay per call, with MCP support. This repo uses:

Each API maps to one function in lib/aisa/sources.py, with paths and parameters checked against AIsa's official OpenAPI spec. Every request carries a price cap and anything over it is rejected. Not sure what a call costs? Add --quote to ask first; quotes are free.

The AIsa links are the author's referral links. The Douyin comments module doesn't use AIsa; it runs in your own logged-in browser for free.

Optional: run logs

For a sturdier setup where you can trace what went wrong, add the run log module. It records:

A nightly health check lists failed, stuck and missed runs, plus cost spikes.

⚠️ It uses more of your bot's quota, so it's off by default and the bot asks during setup. There are three levels:

  • Light: approvals and rule changes only, almost no extra quota.
  • Standard: adds a record for every routine run.
  • Full: adds the nightly check and weekly archiving.

Logs are local files only, so AIsa costs don't change. (The module's doc is in Chinese; agents read it fine.)

Repo layout

saas-growth-agents/
├── AGENTS.md               setup steps for the agent
├── bootstrap.md            the one sentence to send your bot
├── lib/aisa/               AIsa layer: client.py (stdlib), sources.py (data sources), aisa_cli.py (CLI)
├── lib/runlog/             optional: run log, approvals, rule changes, nightly health check (uses more bot quota)
├── bots/
│   ├── newterm/            new-term radar: job.md · routines.md · check.md · pipeline/ · skills/ · install.sh
│   ├── site-ops/           site ops: job.md · routines.md · check.md · scripts/ · skills/
│   └── social-kb/          social analysis pack: README.md · job.md · routines.md · check.md · scripts/ · douyin-comments/
├── extras/trade-radar/     trade radar (extra)
├── skills/agent-setup/     for Claude Code / Codex: asks what you need and writes a tailored setup message for your bot
├── templates/              ledger formats
└── tests/                  dry-run tests that need no key (all sample data is fictional)

Every bot folder has the same shape: job.md is the job description, routines.md the schedule, check.md the acceptance checklist.

Try it without a key

git clone https://github.com/Pluviobyte/saas-growth-agents.git && cd saas-growth-agents
python3 tests/test_dry_run.py                                        # full dry run
AISA_DRY_RUN=1 python3 bots/site-ops/scripts/serp_sites.py "ai video generator"

Boundaries

License

MIT. Third-party services and credits are in NOTICE.md.

saas-growth-agents (Pluviobyte/saas-growth-agents)

Searching Database…

What is this?It is an Agent Skill for Claude Code that scrapes the complete comment section of a single Douyin video (top-level comments plus every nested reply) and compiles them into a readable report.
Is this a standalone scraper app, or does it run inside Claude Code?It is not a standalone app. It runs as a Claude Code skill that drives your own already-logged-in Chrome tab through the Claude in Chrome extension, replaying the page's real requests while the page's own SDK computes the signature.
How do I install and use it?Drop the skill folder into your project's .claude/skills/ directory, then inside Claude Code give it a Douyin video link and tell it to grab the comments; it accepts full video URLs, profile links with modalid=, and v.douyin.com short links.
What do I need installed before this will work?You need Claude Code with the Claude in Chrome extension connected (run /chrome), a Chrome browser already logged into Douyin, plus Python 3 (standard library only) and curl.
What happens when Douyin shows a captcha, and how much can it actually scrape?It will not click through captchas for you — the skill just pauses until you solve the captcha manually in Chrome. It only handles one video at a time (no batch scraping, no Xiaohongshu or Bilibili), and on videos with over ~1000 comments the API starts returning duplicates after about 500, so the real total is sometimes undercounted (823 captured out of 1067 shown in one test).
Who is this meant for?It is built for personal research and content reference on a single video, not for agencies doing bulk collection or commercial redistribution.
Why doesn't it need TikHub or Playwright?Because it simply replays the exact request the Douyin page itself already makes and lets the page's own SDK compute the risk-control signature (abogus, msToken, uifid), so there is no need to pay for a third-party API or reverse-engineer the signing algorithm.
What does it produce at the end?A deduplicated, like-sorted Markdown report (评论全文.md) plus the raw JSON, written by default to .internal/抖音评论/ inside your current project.
Can it download the video too?Yes, downloading the source video at 720p is an optional extra step of the scrape.
How fast is it compared to scrolling manually?In the README's tests it pulled 195 comments from a 7-minute video in 138 seconds (35 requests) and 135 comments from a 46-minute video in 252 seconds, matching results obtained by manually clicking 'expand replies', which took about 6 minutes by hand.
Creator
Pluviobyte
Size
0.4 MB
Stars
542
Updated
CommitNov 10
Age
19d
Commits · 90 Days