Sales outreach · checked 2026-10-19

Can I vibecode ReplyGain?

KINDA· weekend project

The loop is easy for one person: poll a few sources for your keywords, ask an LLM how likely the author is to buy, keep the good ones in a list and draft a reply. An agent builds that in a sitting. It becomes a weekend because of the sources: Reddit's API terms and approval, X's paid API, and the cost and tuning of the LLM scoring. A tool that only watches Hacker News and Bluesky is genuinely free to run; one that also covers Reddit and X is not.

$29/moprice

$348you'd save a year

weekendbuild time

—people replaced it

Before you start

What you need on hand.

DifficultyIntermediate · Some system setup or keys.

TimeAn evening, plus time to get Reddit and Bluesky credentials

Claude Code, Codex or CursorAny of them can run the prompt.
Node.js 22better-sqlite3 compiles a native module, so build tools must be installed.
An LLM endpointA hosted API key, or Ollama on your own machine; scoring is many small calls.
Reddit and Bluesky credentialsReddit asks you to register an app; check its current API terms for your use.
A machine that stays onA small VPS is enough.

Before you buildF5Bot is free for keyword alerts on Reddit, Hacker News and Lobsters, and OpenMagpie is a free open-source scorer. Try them before building.

1. Install one agent: Claude Code, Codex, Cursor.
2. Make an empty folder and open it in the agent.
3. Paste the prompt below, press enter, and answer its questions as it builds.

The prompt

Build me a personal lead monitor to replace ReplyGain. Requirements:

- Node.js 22 + TypeScript: a Fastify server with server-rendered pages (Eta) and a little vanilla
  JS. `npm start` serves http://localhost:4880; a `croner` job runs a scan every SCAN_HOURS (default 4).
- Storage: better-sqlite3 in ./data/leads.db: products (name, url, pitch), keywords (product,
  phrase, kind: intent|competitor|brand), posts (platform, external id unique per platform, author,
  title, body, url, created_at), leads (post, score 0-100, reason, draft_reply, status
  new|replied|skipped, product).
- Sources behind a `Source` interface: Hacker News via the Algolia search_by_date API, Bluesky via
  `@atproto/api` searchPosts with BSKY_HANDLE and BSKY_APP_PASSWORD, Reddit via the OAuth
  client-credentials flow with REDDIT_CLIENT_ID and REDDIT_CLIENT_SECRET (search sorted by new).
  Remember the newest timestamp per keyword and source so a scan fetches only new posts.
- Scoring: send each new post and the product pitch to an OpenAI-compatible endpoint (`openai`
  SDK, OPENAI_BASE_URL and MODEL in .env, so Ollama works) and require JSON with {score, reason}.
  Posts of at least MIN_SCORE (default 60) become leads; only those get a second call that drafts
  a reply in a plain, non-salesy tone, naming the product at most once and only when asked for.
- Cost cap: MAX_LLM_CALLS_PER_DAY in .env. Stop scoring when it is reached and show a notice.
- Pages: a lead feed sorted by score with the post, the reason and an editable draft; buttons Copy,
  Replied and Skip; filters by product and status; a keywords page; an Add product form that
  asks the LLM to suggest 10 keywords from the product's URL, which you edit before saving.
- Alerts: one email digest per day through nodemailer (SMTP in .env) and an optional Telegram
  message per lead with score at least 80 (TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID).
- Export all products, keywords and leads as one JSON file. Never post anything to any platform.
- Single user, one password in .env (signed cookie). No accounts, no telemetry, local-first.
  Outbound calls: the three sources, your LLM endpoint, SMTP and Telegram.
- Out of scope: X (Twitter), LinkedIn, auto-posting, teams, billing.
- README: how to create the Reddit app and the Bluesky app password, a warning to read Reddit's
  current API terms before running this, and a rough LLM cost per 1,000 scored posts.

Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.

Share on X

Max MiskowSponsoredAI agents, automations and MVPs, built for your company.Visit site ›

Then keep going

Next prompts, same folder.

The first prompt builds the core. Run these one at a time in the same folder to grow it into your own ReplyGain.

Improve the scoring in the lead monitor in this folder. Requirements:

- Store a reason when a lead is marked Skip (a short dropdown: not a buyer, wrong product,
  too old, spam, other).
- Include the five most recent Replied examples and five most recent Skip examples with their
  reasons as few-shot examples in each scoring prompt, per product.
- Show on a stats page the share of leads replied to against skipped, per week and per source.
- Update the README.

Add two more sources to the lead monitor in this folder behind the existing Source interface.

- RSS and Atom feeds listed in .env, parsed with `rss-parser`, filtered by the product keywords.
- Mastodon hashtag and search timelines from MASTODON_INSTANCE, public posts only.
- Dedupe across sources by normalised URL and title.
- Add tests with recorded fixtures. Update the README.

Add a weekly insights page to the lead monitor in this folder. Requirements:

- Once a week, send the last 7 days of lead posts for each product to the LLM and ask for
  the five most common problems, each with two quoted examples and their links.
- Store the report, show the history on /insights, and email the newest one.
- Quote only text that exists in the stored posts; drop any quote that doesn't.
- Update the README.

What you lose

X (Twitter) monitoring, which needs a paid API tier of its own
Scans that run around the clock on someone else's servers
Automatic keywords, competitors and communities suggested from your website URL
Telegram alerts with reply and skip buttons
Several websites and long lead history managed for you

Why people still pay

A free keyword alert tool such as F5Bot emails every match, so a founder still reads dozens of irrelevant posts. People pay to have those posts ranked by buying intent, with a reply already drafted, and to avoid maintaining Reddit and X access themselves.

Moatintegrations execution polish

Do the math

ReplyGain or your own copy?

Over

Your usage

Count my time12h ×$/h

ReplyGain, 1 yr$348

Your own copy, 1 yr$98running + build session

You keep$251pays for itself in the first month

Estimate · A local Ollama model costs nothing; a hosted small model for a few thousand posts a month is a few dollars, plus about $5 for a VPS if the machine can't stay on. The build is token usage on your agent plan.

Leaving ReplyGain

Take your data, then cancel.

1 · Take your data out of ReplyGain

1Open Settings, then the Account tab, find Your Data, and press Export JSON.
2The JSON file holds your websites, keywords, platform settings and leads.
3The privacy policy also lists deleting individual monitors or lead records yourself.
4Replies you posted live on Reddit, Hacker News and the other sites under your own accounts, so they need no export.

2 · Cancel ReplyGain

1In Settings under Subscription, press Manage Billing.
2The terms say you can cancel any time and keep access until the end of the billing period.
3Monthly plans: contact ReplyGain within 7 days of the first charge for a full refund. Yearly plans: prorated refund for unused months in the first 30 days. 7-day pass: refund if you find no leads within 3 days.
4Full account deletion is done by contacting ReplyGain, as the privacy policy describes; the Danger Zone tab is the place to look first.

Export the JSON before cancelling; the keywords are the part worth keeping and they work in any other tool.
Plan changes: an upgrade takes effect at once, a downgrade at the start of the next billing cycle.
The account page offers a 3-day Pro trial with no charge today; read the checkout screen before entering a card, and note the date the trial ends.
Replies you posted stay on Reddit, Hacker News and the other sites under your own accounts, so leaving costs you no public history.

Checked 2026-10-19 against ReplyGain's docs: 1, 2, 3, 4, 5, 6

The escape hatch

Don't feel like building it? These are already free.

OpenMagpieOpen sourceApache-2.0 social listening CLI: you describe the posts you want in plain words and a local LLM scores new ones. Docker and uv needed; a hosted version is announced.★ 70 · active Oct 2026 · docker to self-host
CLIVisit ↗HarkenOpen sourceMIT Python tool that follows a keyword across Hacker News, Reddit, Bluesky and others into a local dashboard with sentiment; no intent scoring or reply drafts.★ 37 · active Jul 2026 · one-click install
Web, CLIVisit ↗F5BotFreeFree keyword alerts by email for Reddit, Hacker News and Lobsters; Slack, Discord and feeds are on a paid tier. No scoring or reply drafts.they host it
WebVisit ↗

The numbers

ReplyGain pricing

Plan	Monthly	Annual, per month	What you get
Starter	$29	$22	1 website, 10 keywords, 1,000 AI leads and 1,000 AI replies a month, scans every 12 hours, 90 days of lead history, Reddit, Hacker News, Bluesky and X.
Pro	$79	$59	5 websites, 30 keywords, 5,000 leads and 5,000 replies a month, scans every 4 hours, 180 days of history, custom reply prompts and Telegram alerts.
Business	$199	$149	10 websites, 60 keywords, 10,000 leads and 10,000 replies a month, scans every 2 hours, 365 days of history, custom prompts and Telegram alerts.
7-day pass	custom	—	Full Starter access for 7 days, one per account.

Free tier: No free hosted plan.
Billing: Fees are billed in advance, monthly or yearly, and you can cancel any time with access until the end of the period. Monthly plans have a 7-day money-back guarantee; yearly plans get prorated refunds for unused months within the first 30 days. The pricing page shows the yearly per-month figure by default and has a toggle for monthly billing. Launch discount codes take $10, $30 or $60 off the first six months.
Hidden costs: Leads and replies are monthly quotas. Extra credit packs cost $9 for 100 leads plus 100 replies and never expire. Keywords are counted across all websites.
Source: replygain.com

Also on the list

Apollo.io$65/mo · Sales outreachNOT REALLY

Instantly$47/mo · Sales outreachKINDA

Gojiberry AI$99/mo · Sales outreachNOT REALLY

Questions

ReplyGain, briefly

Not answered here? Tell us and we'll add it.

Ask a question ›

No free plan. A one-time $9 7-day pass gives Starter access, and the account page offers a 3-day Pro trial with no charge.

Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: X (Twitter) monitoring, which needs a paid API tier of its own, Scans that run around the clock on someone else's servers.

About $29 a month on the Starter plan, $348 a year (checked 2026-10-19).

X (Twitter) monitoring, which needs a paid API tier of its own; Scans that run around the clock on someone else's servers; Automatic keywords, competitors and communities suggested from your website URL; Telegram alerts with reply and skip buttons; Several websites and long lead history managed for you. If any of those matter to you, keep paying.