Open Pulse — Open Pulse

Open Pulse watches public conversations and public job postings for the moments that mean somebody needs what you sell, judges every candidate against what you actually do, and delivers what survives by email, signed webhook or MCP.

What this page answers

What is Open Pulse?

A social-listening tool that reads public posts and job postings as buying signals. You describe what you sell; it searches the sources on a schedule, judges each candidate for intent, scores it, and sends you the ones worth reading. It is not a keyword alert: the classifier reads the post against your business rather than matching a word.

What does it watch?

Reddit, Hacker News, X, LinkedIn, Ashby, Greenhouse, Lever, G2, Trustpilot, YouTube, Google Reviews. Only publicly available results are ingested — it does not log in to any platform or bypass access controls.

What kinds of thing does it look for?

Four listener types: Find customers; Situations that mean somebody needs you; Competitor dissatisfaction; Local prospects.

How do results reach me?

Email reports — A summary after every completed run, and a short note when one fails. Recipients resolve most-specific-first: the listener's own list, then the workspace's, then the workspace admins. Webhooks — The same information in a shape another system can act on. HMAC-signed, retried with backoff, and every attempt kept with its request and response so a failure is debuggable from your side. MCP connectors — Open Pulse is an MCP server, so Claude, ChatGPT, Grok and the Claude Code CLI can read your workspace directly. Ask for this week's high-intent signals and get them, in the chat you already work in.

How does it avoid sending me noise?

Cheap gates first, then a reranker, then a classifier that labels intent and scores it. Every rejection is recorded with its reason, so the filter is auditable rather than a black box, and rating a signal calibrates the next pass.

What does it cost?

There is a 14-day free trial with a card up front. See /pricing for the plans.

At a glance

Sources: Reddit, Hacker News, X, LinkedIn, Ashby, Greenhouse, Lever, G2, Trustpilot, YouTube, Google Reviews
Listener types: 4
Trial: 14 days, card up front
Delivery: Email reports, Webhooks, MCP connectors
API: REST under https://api.openpulse.cloud/v1, and an MCP server at https://api.openpulse.cloud/mcp

Sources

Reddit
Hacker News
X
LinkedIn
Ashby
Greenhouse
Lever
G2
Trustpilot
YouTube
Google Reviews

How it works

01. Describe what you sell. Paste your website. Open Pulse crawls it and proposes the vocabulary (keywords, competitors, the phrases your buyers actually use) and you edit it before anything runs.

02. It listens on a schedule. Each run searches the sources for that objective, pulls full post text where the platform allows it, and drops anything older than the listener's window.

03. Every candidate is judged. Cheap gates first, then a reranker, then a classifier that labels intent and scores it. What gets rejected is recorded with the reason, so the filter is auditable rather than a black box.

04. The good ones reach you. A run summary by email, a signed webhook into your own systems, or your agent reading the workspace directly over MCP. Rate a signal and the next pass calibrates.

Listener types

Find customers: People describing the problem you solve, asking for recommendations, or complaining about the workaround they are using instead.
Find customers — sources: Reddit, Hacker News, X, LinkedIn, Facebook
Find customers — window: 30 days
Find customers — labels: high intent, evaluating, relevant
Situations that mean somebody needs you: A company hires three support agents, signs a lease, raises a round, or says their vendor is slow. Each one means something different depending on what you sell, and each is read against your business rather than against a keyword.
Situations that mean somebody needs you — sources: Reddit, Hacker News, X, LinkedIn, Ashby, Greenhouse, Lever, LinkedIn Jobs
Situations that mean somebody needs you — window: 30 days
Situations that mean somebody needs you — labels: they said it, one step away
Competitor dissatisfaction: Rivals are retrieved rather than recalled, then rolled up one row per competitor, so you see who is losing users and why, not a list of mentions.
Competitor dissatisfaction — sources: Reddit posts + comments, Hacker News, YouTube, G2, Trustpilot, X
Competitor dissatisfaction — window: 30 days
Competitor dissatisfaction — labels: switching, frustrated, comparing
Local prospects: Google reviews for the businesses you sell to, themed and scored, so a run tells you which locations are at risk before their next renewal.
Local prospects — sources: Google Reviews
Local prospects — window: 90 days
Local prospects — labels: pain confirmed, at risk

How quality is kept

Full post text where we can get it. Reddit and Hacker News return the body, the author and the engagement. Every source is labelled in the wizard as full text, snippet or best effort, because a classifier judging a search snippet is the single biggest cause of a bad label, and you should know when that is happening.

Rejections are recorded, not discarded. Each dropped candidate keeps its reason. When a listener feels quiet, the quality report shows whether the queries returned nothing or the gates threw everything away.

Per-query and per-source yield. Which searches earn their cost and which never produce a kept signal. Queries that consistently return noise are pruned.

Your thumbs move the bar. Rating a signal feeds the next pass, so the definition of relevant becomes yours rather than the one a model guessed at on day one.

How results leave

Email reports: A summary after every completed run, and a short note when one fails. Recipients resolve most-specific-first: the listener's own list, then the workspace's, then the workspace admins. (Per-run summary, Top signals inline, Failure notices)
Webhooks: The same information in a shape another system can act on. HMAC-signed, retried with backoff, and every attempt kept with its request and response so a failure is debuggable from your side. (signal.created, run.completed, run.failed)
MCP connectors: Open Pulse is an MCP server, so Claude, ChatGPT, Grok and the Claude Code CLI can read your workspace directly. Ask for this week's high-intent signals and get them, in the chat you already work in. (search_signals, get_rollup, get_quality_report)

Frequently asked questions

Where do the signals come from?

Public conversations and public job postings: Reddit, Hacker News, X, LinkedIn and Facebook for market listening; Ashby, Greenhouse, Lever, LinkedIn Jobs and Indeed for postings; G2, Trustpilot, YouTube and Google Reviews for competitor and local work. Only publicly available results are ingested: Open Pulse does not log in to any platform or bypass access controls.

How is this different from a keyword alert?

An alert tells you a word appeared. Open Pulse retrieves the post, judges intent against what you actually sell, scores it, records why it rejected everything else, and learns from the signals you rate. The output is a short list worth reading rather than a feed you stop opening after a week.

What happens after the 14-day trial?

You enter a card to start the trial and nothing is charged during it. The trial then converts to the plan and billing period you chose: $49 a month for Go, $499 for Pro, or $540 and $5,490 a year, and continues until you cancel. You can cancel during the trial from workspace settings and you will not be charged.

What happens if I cancel?

Your subscription stays active until the end of the period you have already paid for, and your listeners keep running until then. After that they stop, and the workspace becomes read-only: every signal, run and report you have collected stays readable for as long as the account exists. Subscribe again and the listeners that were stopped start back up on their original schedules.

Can I connect it to my AI assistant?

Yes. Open Pulse is a Model Context Protocol server, so Claude, ChatGPT, Grok and the Claude Code CLI can read your workspace once connected. The tools are read-only apart from rating a signal and queueing a run, and an agent can never create or delete a listener.

How often does a listener run?

Go runs daily or weekly. Pro adds hourly and 6-hourly. Each run costs retrieval and model spend, which is why the cadence is the difference between the plans rather than an arbitrary lock.

Do you need access to my CRM or my inbox?

No. Open Pulse reads public sources and sends results outward: by email, by signed webhook, or to an AI client you connect. It asks for no credentials to anything you own.

Start the free trial — Creates a workspace. A card is taken up front and nothing is charged during the trial.
Pricing — The two plans, their limits and the annual prices.
Book a demo — A 30-minute call. Faster than the contact form for anything a page cannot settle.
Developers — The REST API, the MCP server and the scopes a key can hold.
Contact sales — Writes to support@openpulse.cloud. The right channel for a security questionnaire or a contract.