Glossary

Every discatch term explained in plain language: social listening, catch, source, quote, and more.

Social listening: Watching public online conversation for mentions of a brand, product, competitor, or topic, and acting on what you find. A company tracks its name on Hacker News, Discourse forums, Reddit, and the web to catch questions, file bugs, and learn what people compare it to. Social listening differs from social monitoring in the why: monitoring counts the mentions, listening asks what to do about them. Tools in the category include Octolens, Brand24, Syften, F5Bot, and Google Alerts.
Discourse: Two meanings, both relevant. (1) Online conversation in general, which is what discatch catches. (2) Discourse, the open-source forum software at discourse.org, which powers thousands of community forums and exposes a public JSON API. discatch treats Discourse forums as a first-class source.
Catch: discatch’s word for a standing alert: a query plus a destination. When new matching discourse appears, the catch fires and delivers it.
Mention: One post, comment, thread, or page that matches a query.
Unit: The smallest thing discatch sells: a search, a capture, a result, an alert delivery, or a classify pass. Agents can buy units one at a time.
Waterfall: Trying providers in order, cheapest and most accurate first, falling through to the next on a miss or failure, and merging results. Borrowed from contact-data products. Nobody in social listening claims it yet.
Provider: A third-party API discatch calls under the hood. Providers are internal and never shown to customers.
Source: A platform where discourse lives: Hacker News, GitHub, Discourse forums, the web, and so on. Sources are what customers care about. Providers are how we reach them.
Classify flag: An optional parameter that asks discatch to judge relevance, sentiment, or intent itself instead of returning raw candidates for the caller’s model to judge. Costs a surcharge sized to the expected LLM cost.
Prescore: A free, non-LLM relevance estimate between 0 and 1 on every result, from exact phrase hits, term coverage, title hits, recency, engagement, and a per-account source weight. It ranks and filters; it does not decide relevance. That is the caller’s model, or the classify flag.
Push-to-agent: Delivering a catch by calling something that wakes an agent: a webhook, a Claude Code or Codex command, a Doozy todo. The opposite of the agent polling an API.
Event log: The per-account, append-only log every catch writes to. Agents read it from a cursor, stream it over SSE, or subscribe targets to it. Webhooks, email, and MCP reads are all consumers of the same log.
Quote: The price of a request before it runs: a ceiling with locked unit prices, valid for a short window. The charge is actual units at those prices and never exceeds the ceiling.
Allowance: Per account and per unit kind, how many units the account may consume in the period. A quote whose ceiling exceeds what remains is refused before anything runs; discatch never truncates and never overdraws.
MCP: Model Context Protocol. The standard way an AI agent connects to a tool server. MCP is pull-only: the agent calls tools; the server does not push. discatch ships an MCP server (planned) and also ships push, because MCP alone cannot.
Agent-first: Designed so an AI agent, acting on behalf of a person, is the primary reader and buyer. Docs, pricing, errors, and offers are written for a model to read and act on.
Doctor: The one entry point: a manifest that tells an agent everything it needs to sign up and make a first call. The same JSON at the API root, at /doctor on the site, from the CLI, and as an MCP tool. Every error’s next_action points at it.
next_action: A field on every error and on the doctor manifest that says what to do next in one sentence, with a link home. Agents act on it; humans read the message beside it.
PLG: Product-led growth. Users start free and self-serve, and pay when they grow. Crisp is the reference discatch follows.
Playbook: A copyable script, mini app, or quick start that plugs into the discatch API to do one thing, such as draft a reply with the user’s own agent. The open-source face of the product. Also called a cookbook.
Buddy: The discatch creature, Beacon, built with buddykit. It reflects product state: listening, working, needs you, joy on a catch.
Dogfood: Using discatch on dcouple’s own products, Doozy, Pane, and Bloom, before asking anyone else to.
Brand monitoring: Tracking where and how often a brand is mentioned online. Often used interchangeably with social listening, but the accepted distinction is: brand monitoring counts mentions and measures sentiment (the what and the now), while social listening asks why people say what they say and what to do next. discatch does both. It catches mentions and pushes them to an agent with a suggested action.
Social listening vs social monitoring: Social monitoring tracks what is said and when: counts, sentiment, alerts on a keyword. Social listening asks why it is said and what to do next: trends, intent, competitor positioning, product feedback loops. Most tools do both under one label. discatch catches the mentions (monitoring) and pushes them to an agent that decides the response (listening).
Keyword alert: A standing notification that fires when new content matching a keyword or phrase appears on a source. Google Alerts is the reference everyone knows. In discatch the keyword alert is a catch: a query plus a destination, re-evaluated on an interval. Catches are not priced per keyword.

Last reviewed 2026-09-10 by Parsa Khazaeepoul, founder, dcouple. Copy from docs/glossary.md.