BLOG-2921 1

A centralized Threat Intelligence Platform (TIP) collects, aggregates and organizes data on known and emerging threats, acting as connective tissue between raw telemetry and active defense. The long-standing problem in the industry is data gravity: teams sit on billions of rows of telemetry while still lacking insights they can act on.

Cloudflare's TIP is built to avoid that trap. It removes the need for complex ETL (Extract, Transform, Load) pipelines by using a sharded, SQLite-backed architecture, and it runs GraphQL directly on the edge so security teams can visualize and automate threat response in real time. Rather than maintaining a single massive database, Threat Events are distributed across thousands of logical shards, which keeps query latency sub-second even when millions of events are aggregated across global datasets.

Combining global telemetry with manual investigations by analysts produces a single source of truth. The platform is intended to carry teams from observing a threat to preemptively blocking it across the Cloudflare network — telling them not only that something is "bad," but why it is happening, who is behind it, and how to prevent it automatically.

A sharded SQLite foundation, queried in parallel

Making every piece of Cloudforce One telemetry actionable required solving a storage problem first: how to serve low-latency, complex queries over billions of events without a centralized database in the middle. The answer was a sharded architecture on SQLite-backed Durable Objects. Each shard is a Durable Object exposing a consistent, transactional interface to its own private SQLite database, and spreading events across a high-cardinality fleet of them keeps any single database from becoming a point of contention during heavy ingestion.

The rest of the stack slots in around that choice. Cloudflare Queues ingest and distribute incoming telemetry asynchronously so attack spikes don't saturate write throughput; R2 holds data for long-term retention while the "hot" index stays in each Durable Object's SQLite storage for instant retrieval.

Reads exploit the distribution rather than fight it. A user query hits the GraphQL endpoint, itself running in a Worker, and the platform fans the request out to multiple Durable Objects in parallel instead of scanning one table. Because those objects are spread across the global network, results aggregate with minimal latency. Permissions are verified first, and shards that cannot hold the requested events are eliminated by date before the fan-out:

// A conceptual look at fanning out a query to multiple shards
async function fetchFromShards(shards, query) {
  const promises = shards.map(shardId => {
    const stub = TELEMETRY_DO.get(shardId);
    return stub.querySQLite(query); // Calling the DO's storage method
  });

  // Parallel execution across the Cloudflare network
  const results = await Promise.all(promises);
  return results.flat();
}

Whether you audit a single dataset across a year of history or synthesize a month of activity across every dataset in the account, the experience stays fluid: the SQL executes where the data lives, so there is no monolithic database to bottleneck on.

From numbers to narratives: graphs, pivots, and filters

diagrams

Spreadsheets don't tell stories; patterns do, so the platform ships dynamic visualizations of the threat landscape. Sankey Diagrams trace the flow of attacks from origin to target, showing which regions are hit hardest and where the underlying infrastructure sits. Industry and dataset distribution views let you pivot instantly to check whether a campaign targets your sector — Finance or Retail, say — or is broad-spectrum commodity activity.

A lone indicator such as an IP address means little without relational and historical context. Threat Insights act as a pivot point that correlates events across datasets into a single campaign or exploit, and the platform maps its internal actor nomenclature to recognized industry aliases — linking internal tracking to names like "Fancy Bear" or "APT28" — so local telemetry is interoperable with global research and intelligence feeds. Hunting itself supports high-cardinality searches across the full dataset — IP addresses, file hashes, domains, JA3 fingerprints — with results typically back in seconds.

Those query parameters can be persisted as saved configurations that act as triggers for the real-time notification engine. When incoming telemetry matches your filters, an alert is pushed to your configured endpoints, moving the workflow from pull-based searching to push-based alerting at ingestion time.

Example saved view and associated actions on the Cloudflare Threat Events Dashboard

Closing the loop: STIX2 exports, one-click rules, and the RFI pipeline

Intelligence is only actionable if the attack surface actually shrinks, so the TIP automates the translation from raw telemetry to enforcement. Organizations running third-party or in-house SIEM and SOAR platforms need interoperability, and mapping internal schemas to STIX2 (Structured Threat Information eXpression) is traditionally a high-latency ETL job. Here the translation happens at the edge: a Worker dynamically maps SQLite records to the STIX2 JSON schema when an export is requested. Raw IP addresses, file hashes, and domain names become standardized STIX cyber observables; relationship objects built from the platform's internal mapping link indicator objects to threat-actor or malware objects, preserving investigation context; and modified and created timestamps are managed automatically in UTC so downstream tools can follow a threat's evolution.

Enforcement is likewise one step from discovery. Spot a malicious pattern in a Sankey diagram or an Actor campaign and you can generate a security rule with a click. The TIP talks directly to the Cloudflare Firewall Rules API, compiling the filtered attributes of your investigation — a specific JA3 fingerprint combined with a list of known malicious ASNs, for example — into a wire-protocol rule deployed across the global network in seconds.

Automation handles the bulk of the telemetry, but the hardest threats still need human intuition. A Requests for Information (RFI) Portal is built into the platform, letting users task Cloudforce One analysts with deep-dive investigations. It is a data-enrichment pipeline rather than a ticketing system: a subscriber spends "tokens" to open a request, and the RFI Worker pulls the relevant Threat Event IDs from the sharded SQLite storage and packages the telemetry for the analyst. Cloudforce One analysts then work in an internal version of the TIP to reverse engineer or pivot across global datasets. Their findings — new IOCs, actor attributions, campaign notes — are written back into the global intelligence feed. Metadata from a completed investigation is pushed back to the edge as a threat event where relevant, ready for the WAF or Firewall rules already configured: human analysis feeds the platform's automated detection logic in real time, replacing the static "report" model with intel-as-code.

Request for Information dashboard example, tracking usage of tokens

The edge as the compute layer for threat intelligence

Pushing storage, aggregation and visualization onto the Cloudflare global network changes where the work happens, not just which screen the analyst looks at. The "data gravity" that normally slows a SOC down is removed: queries no longer wait for logs to replicate into a central repository. Questions like whether an IP is tied to a known campaign are answered in the same environment that filters the traffic.

Three tiers of Cloudforce One access

Access is split into three functional levels so that the offering scales with team size and requirements:

  • Cloudforce One Essentials — default datasets in threat events, indicator search, and threat hunting investigations.
  • Cloudforce One Advantage — adds Threat Intelligence Analyst custom insights delivered through requests for information.
  • Cloudforce One Elite — the complete package: brand protection, a high number of requests for information, and access to all threat events datasets.

Consolidation over pipeline maintenance

Adversary infrastructure evolves faster than the Internet it runs on. Keeping telemetry and response logic inside a single integrated platform lets defenders spend their effort on hunting rather than on building and maintaining pipelines.