Lead Signal Listener

Find people who are publicly describing the problem you solve — before they find your competitors.

Lead Signal Listener is a local, open-source lead discovery tool. It listens to Reddit, X, LinkedIn, Facebook, Instagram, Hacker News, YouTube, news, RSS feeds, and any of the 3,800+ endpoints in the Treg catalog. It also picks up trigger signals such as hiring and funding. Every lead is scored, de-duplicated and queued, so you can review the best ones first.

中文说明 · Agent setup guide · Showcase: built in one AI conversation

Built by newmax.ai · Powered by treg.to

_{The leads screenshot shows real results from one run of the default keywords (284 leads, $0.073 in Treg calls), with personal usernames blurred. The listeners screenshot uses a demo configuration.}

Copy this into Claude Code, Codex, Cursor, NewMax or any coding agent:

Read AGENT_SETUP.md from https://github.com/yangyixxxx/lead-signal-listener
and set up Lead Signal Listener for me. Walk me through choosing keywords and scoring rules.

The agent will install it, interview you about your buyer, propose keywords and scoring rules, run a first pass within your budget, and tune the results with you.

Requires Python 3.10+. No third-party dependencies.

git clone https://github.com/yangyixxxx/lead-signal-listener.git
cd lead-signal-listener
python app.py            # opens http://127.0.0.1:8787  (Windows: double-click start.bat)

1. Settings → paste your Treg token → Save → Verify
2. Listeners → describe who you're looking for, enable sources and add keywords (one per line)
3. Leads → Run now, review by score, set a status, add notes, export CSV/JSON

No token? The free sources still work: Hacker News, RSS and import.

The default keywords and scoring rules are only an example (finding companies that struggle with AI search visibility). Replace them with your own: any audience, any problem. The Leads page describes whatever you configure.

The UI defaults to English. Use the EN / 中文 toggle in the top-right corner to switch.

Find endpoints for custom sources without a token:

python app.py --catalog "youtube search videos"

You can also search the catalog in the UI (Listeners → Treg catalog). Clicking a result adds it as a custom source.

Prices are from the Treg catalog (Oct 2026). The X-Treg-Cost-Micro response header is authoritative. Each keyword fetches up to max_pages pages per run (default 3 for Reddit/X, 2 for LinkedIn; one page = one call). A run of the default config is about 50 calls, roughly $0.07.

• Per-call cap, enforced by Treg through X-Treg-Route-Max-Cost
• Per-run budget: paging and remaining paid calls stop once it's reached
• Idempotent replay: re-running the same query within an hour is not billed again
• Auto-stop: an invalid token (401) or empty balance (402) cancels the rest of the run
• Spend and call counts are shown on the Leads dashboard

The scoring is transparent, rule-based and free. Each lead is matched on:

• pain phrases ("traffic dropped", "doesn't mention us")
• intent phrases ("looking for", "recommend an agency")
• persona words (e.g. SaaS vs. DTC)
• engagement, and whether it's a question
• negative phrases (self-promotion, job posts, spam)

Edit the rules as JSON in Listeners → Scoring rules, then click Rescore all leads to apply them to existing leads. For deeper qualification, export JSON and pass it to an LLM.

python app.py --port 9000                   # different port
python app.py --run-once                    # headless run (cron / Task Scheduler)
python app.py --data ~/listener-data        # custom data directory
python app.py --catalog "tiktok search"     # search the Treg catalog
python app.py --import leads.json           # import leads
TREG_TOKEN=... python app.py                # token from env, never written to disk

The server listens on 127.0.0.1 only. Write requests must send Content-Type: application/json.

GET /api/config · POST /api/config · POST /api/test-token · POST /api/run · GET /api/status · GET /api/signals · GET /api/signals/counts · POST /api/signals/<id> · POST /api/import · POST /api/rescore · GET /api/catalog?q= · GET /api/export?fmt=csv|json

See AGENT_SETUP.md for details.

• Everything stays on your machine: data/config.json (settings and token) and data/signals.db (SQLite).
• data/ is git-ignored. Never commit or share it.
• Server binds to 127.0.0.1 and checks the Host header (blocks DNS rebinding). Writes require JSON.
• The token is masked in the UI and never logged. Team (identity) tokens automatically send X-Treg-Org.

Only collect public content, and respect the terms of each platform and Treg provider. This is a research and lead discovery tool, not a spam cannon. Follow GDPR, CAN-SPAM and similar rules when you reach out.

app.py                  entry point (web UI, --run-once, --catalog, --import)
listener/
  treg_client.py        Treg HTTP client (cost, caps, idempotency, org header)
  sources.py            built-in sources + parsers
  generic.py            RSS/Atom, custom Treg endpoints, import, catalog search
  scoring.py            rule-based scoring
  store.py              SQLite storage + de-duplication + export
  runner.py             one listening pass
  server.py             local JSON API + scheduler
  i18n.py               server-side messages (en / zh)
  static/index.html     single-page UI (en / zh)
  static/fonts/         bundled OFL fonts (no Google Fonts dependency)
docs/                   screenshots
tests/                  unit tests (python -m unittest discover tests)

MIT

• Built by newmax.ai — made by Yangyi · X: @yangyi
• Powered by treg.to — built by Jason Zhou · X: @jasonzhou1993

If this project helps you, a ⭐ on GitHub and a follow on X are much appreciated.