Perplexity on November 16, 2026 launched the Monitor API — a scheduled web and page watcher that fires signed webhooks when its Decider model thinks something changed that matches your goal. It uses the same API key as Search, Context, and the Agent API, and prices at $1 per 1,000 successful runs for topic monitors (page monitors bill per URL fetched).
For builders running persistent agents or custom token gateways, Monitor is the missing push layer: instead of cron-searching competitor pricing, your harness receives a confidence-scored payload when a page diff or news hit passes four checks.
TL;DR
| Question | Answer |
|---|---|
| What shipped? | Monitor API with web (Search-backed) and page (diff) modes |
| Delivery? | Signed webhooks with retries; optional pause notifications |
| Scheduling? | Intervals (15m to weekly) or cron; dedup across runs |
| Pricing? | $1 / 1k successful web runs; $1 / 1k page URLs fetched |
| Key reuse? | Same key as Search / Context / Agent API |
Why agents need push, not only pull
Most agent stacks still implement monitoring as cron + Search API + custom dedup. That pattern breaks when you track hundreds of accounts, regulatory pages, or competitor changelogs — exactly the failure mode Together AI described when agents misread static CLI examples as live inventory.
Perplexity's pitch is to outsource the schedule, deduplication, and relevance gate while your agent keeps write access to CRMs, repos, or loop runners. The launch post lists GTM engineering, equity research, competitive intelligence, and regulatory monitoring as production shapes — all webhook-first workflows.
How Monitor works: targets, Decider, and schedules
Perplexity defines each monitor in plain language:
- Targets — either one search query (web monitor) or 1–10 page URLs (page monitor).
- Goal (optional) — natural-language filter, updatable without recreating the monitor.
- Schedule — fixed interval or cron; runs resume state between executions.
- Webhook — HTTPS endpoint receiving signed events.
Web monitors pipe the query through Perplexity Search across web and news. Page monitors store a baseline, then return text diffs on later runs.
Before delivery, Perplexity Decider applies four thresholds on web results — relevance, source legitimacy, new information, and freshness — plus a meaningful-change check for page diffs. Each passing update includes a confidence score, and thresholds are adjustable per check.
Webhook bell icon illustrating Perplexity Monitor API push alerts for agent workflows
That design matters for agent security: a naive RSS scraper will spam your harness with SEO junk. Decider is Perplexity's attempt to cap webhook volume before your MCP tools act on bad intel.
Pricing and what counts as a run
Perplexity's pricing section states:
- Web monitors: $1 per 1,000 successful runs.
- Page monitors: $1 per 1,000 URLs fetched, because one scheduled run may pull multiple pages.
A run is one scheduled or on-demand check, not one webhook delivery. Failed runs that find nothing may still consume quota depending on console billing; confirm in your API Console usage tab before wiring production agents.
Compare that to always-on Teammates on Max, where usage credits blend model, Computer, and tool spend. Monitor is metered infrastructure for backends, closer to Hugging Face plus OpenRouter routing than to chat subscriptions.
Getting started: curl and webhook verification
Perplexity documents this POST pattern (abbreviated here; see the official post for the full JSON):
curl --request POST https://api.perplexity.ai/v1/monitors \
--header "Authorization: Bearer $PERPLEXITY_API_KEY" \
--header "Content-Type: application/json" \
--data '{
"name": "AI infra funding",
"targets": [
{"type": "search", "query": "AI infrastructure funding rounds"}
],
"goal": "New Series A or B announcements",
"schedule": {"type": "interval", "interval": "1d"},
"webhook": {"url": "https://agent.example.com/hooks"}
}'
The Monitor API reference covers PATCH updates, webhook secret rotation, and default event types monitor.changes.detected and monitor.paused.
Operational checklist for agent teams:
- Terminate TLS on a public HTTPS endpoint; self-signed certs will fail delivery.
- Store webhook secrets to verify signatures before your agent calls write tools.
- Route metadata fields to the correct tenant if one gateway serves many users — the same pattern we describe for multi-user AgentCore memory.
- Rate-limit downstream tools so a burst of regulatory diffs cannot open unbounded shell access.
Perplexity's evaluation claim
The launch post says Perplexity tested Monitor against other Search API providers on 60 production-style topics for eight hours, with a separate model judging relevance and publication timing. It claims more fresh, relevant hits than rivals on many topics.
explainx.ai has not reproduced that study. Treat it as vendor methodology until Perplexity publishes task lists or open logs. For your own stack, run a shadow period: log Monitor webhooks alongside your legacy cron job for two weeks and measure false positives before you grant agents auto-commit rights.
What is claimed vs verified on the $1B run rate
The same news cycle bundled Monitor with a $1 billion annual revenue run rate figure attributed to CEO Aravind Srinivas speaking to Semafor at a live event, as recapped in trade press on November 16–17, 2026. Earlier reporting had cited under $250M at the start of 2026 and over $750M by August, per those recaps.
Verified today: Monitor's product specs, pricing, and API endpoint on Perplexity's own hub and docs.
Not verified in primary filings: The $1B run rate itself. Perplexity is private; there is no SEC filing matching the live-event quote at publish time. If you cite the number, attribute it to Semafor's event coverage, not to audited results.
That separation matters because Monitor pricing is usage-linear. A jump to $1B run rate would imply massive Search, Agent, and Computer volume — but this post is about whether your webhook architecture is worth $1 per thousand checks, not about valuing the company.
Monitor vs Teammates vs roll-your-own cron
| Surface | Best for | Memory | Delivery |
|---|---|---|---|
| Monitor API | Backend triggers, compliance watches | Stateless per monitor | Webhook |
| Teammates | Long-running Max user workflows | Project history in Computer | In-product plus Slack |
| Cron plus Search | Full control, lowest vendor lock-in | You implement | Polling |
If you already run Claude Code loops that grep changelogs hourly, Monitor trades engineering time for Decider plus dedup. If you need human approval on every outbound email, keep the human step — webhooks should enqueue drafts, not sends.
Webhook payloads and handler design
When a run finds relevant updates, Perplexity POSTs to your endpoint with either search hits (title, snippet, confidence) or a page diff. The launch post shows metadata fields so you can route events to the correct workspace, user, or agent session — critical if one backend serves multi-tenant customers.
Perplexity also documents pause events when a monitor stops firing, which prevents silent failure when quotas exhaust or goals become unsatisfiable. Your orchestrator should alert humans when monitor.paused arrives, not only when monitor.changes.detected fires. On November 17, 2026 Perplexity also opened the multi-model Orchestrator API for developers who want the same routing layer outside the Computer UI.
Signed delivery means you should verify HMAC or shared-secret headers before any tool calls — the same hygiene we recommend when wiring FFmpeg skills into MCP servers that accept untrusted filenames. Treat webhook JSON like user input: schema-validate, then pass sanitized snippets into the model context.
When Monitor beats rolling your own cron
| Pattern | Build yourself | Use Monitor |
|---|---|---|
| Five competitor pages | GitHub Action plus diff | Often cheaper in eng time |
| 200 topic watches | Custom dedup plus search bills | Monitor dedup plus Decider |
| Air-gapped compliance | Must stay on-prem | Not a fit — keep internal crawlers |
| Sub-second latency | Webhooks are schedule-bound | Monitor is not a tick data feed |
If your team already runs Bedrock AgentCore with EventBridge, you can fan Monitor webhooks into the same bus your human approvals use. That keeps one audit trail for what triggered an agent, whether the source was cron or Perplexity.
Failure modes to plan for
Goal drift: Updating the goal changes what fires without recreating the monitor — good for product, risky if your agent parsed the old goal into a policy file. Version your webhook handler.
Page monitor blind spots: Dynamic single-page apps may produce noisy diffs. Pair page monitors with structural tests in CI, not only text diff alerts.
Webhook retries: Perplexity promises retries on transient failures. Your endpoint must be idempotent — duplicate delivery should not double-charge customers or open duplicate tickets.
Compliance: Regulatory monitors still need human review for trading or medical decisions. See Apple health-data guardrails for how even strong vendors separate alerting from advice.
Questions teams ask before turning Monitor on in prod
Is Monitor a replacement for Perplexity Search in agents? No. Search answers now; Monitor answers what changed since last Tuesday. Most production stacks need both: Search inside the agent loop, Monitor outside it.
Can I watch paywalled pages? Page monitors only fetch URLs your key can retrieve. Login-gated competitor dashboards still need authenticated crawlers you control.
Does Monitor respect robots.txt? Perplexity does not spell out robots behavior in the launch post. If compliance matters, run a legal review on target domains before pointing monitors at third-party sites.
Will this show up in consumer Perplexity? The API ships to developers first. Teammates users may never see a Monitor UI unless Perplexity surfaces templates inside Computer later.
For hands-on onboarding, explainx.ai runs live workshops on agent harness design — useful when you are deciding whether webhooks should trigger write tools or human-in-the-loop tickets only.
Related on explainx.ai
- Perplexity Teammates on Computer — persistent agents vs headless Monitor
- Nasiko token gateway — routing spend when webhooks fire often
- Loop engineering with coding agents — scheduling agent work safely
- What is MCP — wiring webhook handlers to tools
- Bedrock AgentCore multi-user memory — tenant routing for shared webhooks
- Hugging Face OpenRouter inference — alternative API routing patterns
Official docs: Monitor overview, API reference, launch post.
Specs and pricing are accurate as of November 17, 2026 UTC.
