SignalPipe is an agentic sales pipeline that monitors Reddit, Hacker News, and RSS feeds 24/7 for buying-intent signals. It uses AI to score each signal, draft a reply, and surface only real leads for human review and approval before anything is sent.

How does SignalPipe detect buying intent?

SignalPipe uses a three-stage pipeline: a keyword gate for cheap pre-filtering, OpenAI text-embedding-3-small for semantic similarity scoring against your anchor sentences, and a three-model LLM swarm (Optimist, Analyst, Skeptic) that evaluates each post independently. A hard skeptic veto suppresses any signal the most skeptical judge dismisses.

Does SignalPipe send messages automatically?

No. Every message requires human approval before it is sent. SignalPipe drafts the reply and queues it for your review — you approve, edit, or reject each one. Nothing is sent without your explicit sign-off.

What platforms does SignalPipe monitor?

SignalPipe monitors Reddit (via subreddit RSS feeds), Hacker News (via hnrss.org keyword feeds), and any RSS-compatible feed. You can add as many sources as you want per product.

How is SignalPipe different from Clay or Apollo?

Clay enriches contact lists you already have. Apollo is a cold email database. SignalPipe generates net-new leads by detecting live buying signals — finding people who are actively posting that they need what you sell, right now, in public communities.

SignalPipe has four tiers: BYOK ($19/mo — bring your own LLM key, 3 products, 50 leads/day, 1,000 prospects), Starter ($49/mo — 2 products, 25 leads/day, 500 prospects), Growth ($149/mo — 10 products, 250 leads/day, 5,000 prospects), and Scale ($499/mo — unlimited everything). Open Source is free for self-hosters. The first 100 waitlist members get 50% off for 3 months.

Does SignalPipe work with NVIDIA NeMoClaw?

Yes — fully compatible out of the box. NeMoClaw is NVIDIA's enterprise distribution of OpenClaw, so any OpenClaw plugin installs and runs on NeMoClaw without modification. For on-prem deployments, SignalPipe also supports local model inference: set COMPANION_LLM_PROVIDER=openai_compatible and point it at a locally running Nemotron or any OpenAI-compatible endpoint. Your leads, prospect data, and generated messages never leave your infrastructure.

Can I run SignalPipe fully on-premises?

Yes. Self-host the FastAPI backend on your own server, connect it to your own PostgreSQL database, and configure the companion message generation to use a local model via any OpenAI-compatible endpoint (Nemotron via NeMoClaw, Ollama, LM Studio, vLLM). The only external call that remains is the OpenAI embedding API for signal scoring — and that can be swapped for a local embedding model by request.

OpenClaw Plugin

Add Buyer Intent Detection to Your OpenClaw Agent

Not every community post is a lead. SignalPipe's 4-stage intent pipeline — keyword gate, embedding similarity, sarcasm filter, LLM swarm — finds the 15% that are.

Install SignalPipe on OpenClaw

openclaw plugins install signalpipe

Available on ClawHub · MIT licensed · Open source on GitHub

What is OpenClaw Buyer Intent Detection?

Buyer intent detection is the core challenge in community-based lead generation. Most posts are noise. SignalPipe's pipeline was designed to find the minority of posts that indicate genuine, near-term purchase intent — without flooding your queue with false positives.

How SignalPipe adds Buyer Intent Detection to OpenClaw

The 4-stage pipeline: (1) keyword gate blocks obvious noise, (2) embedding similarity against your anchor sentences catches semantic matches, (3) sarcasm detection removes misleading positives, (4) a three-judge LLM swarm (Skeptic, Analyst, Optimist) votes on each signal. The Skeptic has a hard veto. About 85% of posts are filtered — only the top 15% reach your OpenClaw session.

How to set up Buyer Intent Detection in OpenClaw

Define buyer phrases

Write anchor sentences from the buyer's perspective — phrases that describe the problem your product solves. These train the embedding similarity stage.

Add keyword signals

Add buy_signal_keywords that indicate purchase intent in your category. These feed the keyword gate and boost scoring for relevant posts.

Tune with feedback

Approve and reject leads in your queue. The RL system learns from your decisions and adjusts scoring weights per-product over time.

Review only what matters

After a week of training, your queue contains only high-confidence buying signals. Reviewing 5-10 leads per day replaces hours of manual scanning.

Frequently asked questions

How accurate is SignalPipe's intent detection?

After two weeks of feedback training, most users see false positive rates below 10% in their review queue. The RL system continuously improves as you approve and reject leads.

What makes the LLM swarm better than a single model?

A single model can be confidently wrong. The Skeptic-Analyst-Optimist voting structure forces adversarial evaluation of each post. The Skeptic's hard veto means overly optimistic scores never reach your queue.

Can I adjust the scoring threshold?

Yes. You can configure the minimum score threshold per product. Higher thresholds mean fewer but higher-confidence leads. Lower thresholds mean more volume with more noise.

Does it detect competitor mentions?

Yes. Competitor mention detection is built in. Posts mentioning your competitors are automatically flagged and score-boosted — these convert at the highest rate.

Add Buyer Intent Detection to your OpenClaw agent

Join the waitlist. First 100 get Starter free for 3 months.

No credit card. No commitment.

More OpenClaw use cases

Full setup guide →

All 11 tools, anchor sentences, station types

How intent scoring works →

4-stage pipeline explained

OpenClaw Sales Agent →

The first intent-based sales plugin for OpenClaw

OpenClaw Lead Generation →

Stop building cold lists

OpenClaw Community Prospecting →

Community prospecting finds buyers before they find your competitors

OpenClaw Outreach Automation →

SignalPipe drafts replies to community posts so you can respond in 30 seconds