Built & orchestrated by one human
Pulse runs a continuous AI pipeline — ingesting from dozens of sources around the clock, enriching every article through a multi-agent chain, applying editorial quality gates, and delivering curated morning briefings. No team. No editorial staff. Just 12 agents — 10 of them LLM-powered.
Live metrics active
6 stages · 12 agents · 10 LLM-powered
Every piece of intelligence travels through 6 stages before reaching a subscriber's inbox. 10 agents are driven by large language models; the rest by deterministic automation.
Scheduled automation
Polls RSS and web sources on a continuous schedule, normalizes raw entries, deduplicates, and stages new articles for enrichment.
Cron + rule-based automation
Orchestrates source-specific scrapers and cron-driven refresh cycles. Detects stale sources and triggers re-fetch. Runs independent of the AI pipeline.
LLM (base)
Parses article HTMLs and non-standard RSS feeds and prose to extract all referenced URLs, author bylines, and publication metadata.
LLM (base)
Filters extracted URLs to primary sources only — vendor advisories, official research, major wire services. Strips noise and aggregator links.
LLM (base)
Generates an impact-led summary, assigns threat/relevance keywords, and scores each article for editorial priority. Retries on parse failure.
Rule-based + LLM scoring
Evaluates enriched articles against quality thresholds, applies status transitions (pending → reviewed → published), and flags low-signal items for human triage.
LLM (high)
Synthesizes ranked article sets into period-level buzz summaries. Weights by content type (news, intel, exploits, scholarly). Output feeds the Period Chronicle used by the digest pipeline.
LLM (high)
Assembles multi-day narrative rollups (weekly, monthly, quarterly) by chaining buzz summaries into coherent continuity products.
LLM (high)
Selects articles from the reviewed pool, identifies the lead story, and drafts the editorial thesis — grounding the digest in the most important narrative of the day. Uses the Period Chronicle to ensure longer-cadence coverage.
LLM (high)
Drafts all digest sections — situation briefing, context synthesis, what-you-missed, and bullets — using the curator thesis as the editorial anchor.
LLM (base → high on escalation)
Applies quality gates: length policy, verified claim rate, topic deduplication. Escalates to the high-capability model if the draft fails thresholds. Can trigger a full rescue pass.
LLM (base)
Defangs all network indicators of compromise (IPs, domains, URLs) embedded in prose — converting them to non-clickable format to prevent accidental activation in email clients.
24-hour snapshot
Aggregate output across all pipeline stages in the last 24 hours.
Cost analysis
Estimated agent-hours are calculated from workload counts multiplied by fixed effort coefficients. Automation cycles are estimated from unique source refresh activity per UTC day. Labor-cost equivalency converts role-level average annual salaries into productive-hour rates. These values represent planning estimates, not wall-clock runtime.
Assumes 6 productive hours/day per person, Monday–Friday only.
6 / 12 active
in last 24h (7 / 12 over 7d)
3.1h
60.8h over 7 days
1 people
for equivalent daily output (3 for weekly, 5-day week)
$3,861
per week ($195 / day)
11 days
for one person to replicate the 7-day output
1
management role