Skip to main content
Nephia is the monitoring API for brands, names and markets across live Sources (X, Reddit, Youtube, Bluesky, Hacker News, RSS, AI answers — and Vinted). You create a Query: search criteria across the Sources you switch on, polled on an interval. Nephia keeps what matches, reads each mention for sentiment and intent, sorts it into your buckets, and answers your own agent step over it. Read the result back with GET /v1/mentions — free, across every Source and every Query — or have it POSTed to you as a signed webhook event. One search can carry several terms at once — a brand plus its variants — OR’d into a single upstream request at one tick’s cost. For a one-off backfill of what a Source already holds, create a Watch with backfill: true (see Watches).

Sources

Each Source has its own coverage, Watch criteria and event shape. See X, Reddit, Youtube, Bluesky, Hacker News, RSS & News, AI answers, and Vinted. Watching a clothing brand or sourcing resale? Vinted is a first-class Source, with observed price history and market statistics on top of its Watches.

What the API sells

Queries and their mentionsPOST /v1/queries sets the monitoring up, GET /v1/mentions reads what it caught, annotated. Reading your own mentions is free; a Query’s poll ticks are what deduct credits, see Credits. Watches — a Query’s unit of polling, and creatable on their own when a whole Query is more than you need. Nephia polls on an interval and POSTs webhook events when matches change, and the events stay readable from the Watch’s own feed. See Watches. Alongside them: webhooks, POST /v1/analyses over what your Queries collected, AI-answer runs, and Vinted’s observed price history and market statistics.

History & backfill

A Query starts recording the moment it exists, and by default it starts empty: the first poll of each Source records what is there without emitting it, and everything after that is news. Two things change that. backfill: true on POST /v1/queries runs one immediate search per enabled Source and stores the results as the Query’s starting history, so GET /v1/queries/{id}/results answers with content straight away. It costs one tick per Source at that Source’s rate (see Credits), it notifies nobody — no webhook, no channel — and the rows come back with "seeded": true. The default is false: an API default never spends. The same option on a single Watch is documented under Watches. History beyond the first page is a dashboard gesture. On the Sources whose upstream can be asked for older items — Hacker News, Reddit, Bluesky, X, GitHub, YouTube, Stack Overflow, Lemmy and Mastodon — a Query can be handed its last 60 days, read one page at a time up to five pages per Source, each page charged one tick. RSS feeds, AI answers and Vinted listings are not swept: a feed serves what it serves, an answer engine has no archive, and old listings are inventory rather than mentions. Seeded rows never trigger a webhook, a channel, a rule or an agent step, and never count toward a spike alert.

Auth

All product API calls use an API key tied to your Account (billing tenant). See Authentication.

Limits

What a plan sells is queries × terms × freshness: how many things you watch, how many terms each search carries, and how often we check (Limits). Credits are the meter underneath — what each call and each tick costs (Credits) — plus the per-Account request quota on /v1/*. Credit packs top the meter up and never change a limit.