GitHub - thomasbln/openclaw-marketing-agent: Market Signal Radar: AI Agent blueprint to detect Problem-Market Drift using OpenClaw, Node.js and Postgres.

This OpenClaw AI Agent blueprint brings Problem–Market Observability to life: it scans Hacker News, Google News, and X.com for market signals, analyzes them via LLM, and delivers daily alignment digests to Telegram.

📖 Read the article: Detecting Problem–Market Drift Before Your Metrics Break

What it does

• Signal Radar: Multi-source scanning (Hacker News, Google News, X.com) for friction signals (churn, conversion, complaints).
• OpenClaw Agent: Intelligent classification using LLMs (Claude/GPT) against your product's core messaging.
• Strategic Insights: Automated Problem–Market Alignment reports (daily digest, trend comparison, language gaps).
• Telegram Interface: Command-driven access to /trend, /brief, /problem, and /wording 7d.

Quick Start

1. Clone the repo.
2. Setup Environment: Copy the .env.example files in the subfolders and add your API keys.
3. No Secrets: This repo contains no secrets or pre-configured keys.
4. Deploy: Use the provided docker-compose files to spin up the Postgres DB and the Agent. See detailed steps below.

Fork & Adapt

This is a demo and blueprint agent. It is functional out of the box but designed to be extended. Feel free to fork, adapt, and improve it for your own specific use case. If you build something cool on top of this or find a better way to detect drift, pull requests and feedback are always welcome.

Architecture

┌─────────────────┐     ┌──────────────────┐     ┌─────────────┐
│  signal-radar   │────▶│  Postgres        │◀────│  OpenClaw   │
│  (HN, GNews, X) │     │  radar           │     │  + Plugin   │
└─────────────────┘     └──────────────────┘     └──────┬──────┘
        ▲                           ▲                    │
        │ Cron                      │ SQL                │ Telegram
        │                           │                    ▼
┌───────┴───────────┐     ┌─────────┴────────┐   ┌──────────────┐
│ trigger-radar-    │     │ market-radar     │   │  You (DM)    │
│ digest.sh         │     │ plugin (tools)   │   │  or Channel  │
└───────────────────┘     └──────────────────┘   └──────────────┘

Stack: Node.js, PostgreSQL (pgvector), Docker, OpenClaw.
Sources: Hacker News, Google News RSS, X.com.

How the bot works

1. You send a message to the Telegram bot (e.g. "Latest signals", "Brief 7d", "Start now").
2. OpenClaw receives it and routes to the market-radar skill.
3. The skill maps your phrase to a tool:
   • get_signals – Fetch signals from the DB (filter by days, keyword, source)
   • get_source_stats – Counts per source
   • get_trend – Compare two time periods
   • run_scanner – Trigger a new scan (requires elevated permission)
4. The plugin runs the tool (SQL or Docker) and returns data.
5. The LLM (Claude/GPT) formats the response and sends it back to you.

Example flow: "Start now" → run_scanner → plugin starts signal-radar container → "Scanner started. Results in a few minutes."

Components

Prerequisites

• Docker & Docker Compose
• Anthropic API key (Claude) – OpenAI key optional for rate-limit fallback
• Telegram Bot Token (@BotFather)
• ~10 min setup

Quick Start

1. Network & Postgres

docker network create radar-net
cd postgres && docker-compose up -d

Connect to Postgres and init the DB:

CREATE DATABASE radar;
CREATE USER radar_user WITH PASSWORD 'your_secure_password';
GRANT ALL ON DATABASE radar TO radar_user;
\c radar
CREATE TABLE runs (id SERIAL PRIMARY KEY, source TEXT, created_at TIMESTAMPTZ DEFAULT NOW());
CREATE TABLE signals (id SERIAL PRIMARY KEY, run_id INT REFERENCES runs(id), source TEXT, title TEXT, content TEXT, url TEXT UNIQUE, subreddit TEXT);
CREATE TABLE messaging (page_id TEXT PRIMARY KEY, headline TEXT, subheadline TEXT, pain_block TEXT, cta TEXT, version TEXT, updated_at TIMESTAMPTZ DEFAULT NOW());
GRANT ALL ON runs, signals, messaging TO radar_user;
GRANT USAGE, SELECT ON SEQUENCE runs_id_seq, signals_id_seq TO radar_user;

2. Signal Radar

cd signal-radar
cp .env.example .env
# Edit .env: DATABASE_URL=postgresql://radar_user:your_secure_password@pgvector-db:5432/radar
docker-compose build
docker-compose run --rm signal-radar   # Test run

3. OpenClaw

cd openclaw
cp .env.example .env
# Edit .env – see "Env variables" below
mkdir -p config
cp ../openclaw.example.json config/openclaw.json
# config uses ${VAR} – values come from .env
cd plugins/market-radar && npm install && cd ../..
docker-compose up -d

3b. Messaging import (for /wording 7d)

cd openclaw
./scripts/trigger-import-messaging.sh   # Import messaging/homepage.md → DB

4. Auth profiles (API keys)

OpenClaw needs API keys in auth-profiles.json. Either run openclaw onboard interactively, or add them via env: ANTHROPIC_API_KEY and OPENAI_API_KEY in .env are picked up if auth profiles reference them.

5. Cron (optional – daily digest)

crontab -e
# Add: 0 8 * * * cd /path/to/openclaw && ./scripts/trigger-radar-digest.sh

Env variables

Set these in openclaw/.env (no secrets in the repo):

openclaw.example.json uses ${TELEGRAM_BOT_TOKEN}, ${OPENCLAW_HOOKS_TOKEN}, ${OPENCLAW_GATEWAY_TOKEN}, ${TELEGRAM_USER_ID} – all resolved from .env.

Bot commands & phrases

Note: /wording 7d requires get_messaging – add your landing-page texts in openclaw/messaging/*.md, run import, and ensure get_messaging is in tools.allow.

run_scanner ("Start now")

To let the bot trigger scans via Telegram:

1. Docker socket – docker-compose already mounts /var/run/docker.sock and group_add: "docker". If you get "Unable to find group docker", use your socket's GID: group_add: ["1001"] (see ls -la /var/run/docker.sock).
2. Config – openclaw.example.json already has run_scanner in tools.allow and tools.elevated.allowFrom.telegram → ${TELEGRAM_USER_ID}. Set TELEGRAM_USER_ID in .env.
3. Image – Ensure signal-radar is built: cd signal-radar && docker-compose build.

File layout

.
├── openclaw.example.json   # Config template (uses ${VAR} from .env)
├── openclaw/
│   ├── config/             # openclaw.json lives here (not committed)
│   ├── messaging/          # Landing-page texts for /wording 7d (homepage.md etc.)
│   ├── plugins/market-radar/
│   ├── scripts/            # trigger-radar-digest.sh, import-messaging.js, trigger-import-messaging.sh
│   └── workspace/skills/market-radar/
├── signal-radar/
│   ├── sources/            # hackernews, google-news, x (X.com/Twitter)
│   └── keywords/           # EN only: hackernews.json, google-news.json, x.json
└── postgres/

Never commit: .env, openclaw/config/openclaw.json, openclaw/config/agents/. Use .env.example and openclaw.example.json as templates.

Troubleshooting

License

MIT

crafted by thomas @ awareo.io