Sign in

New platform architecture

The speed of cloud analytics.
The control of self-hosting.

Countly 26.01 is one deployable system: streaming ingestion, one analytical store, and analytics, engagement and AI reading from it live. Fast answers, full data ownership, no second vendor.

Talk to a Countly expert

What changed in 26.01

Three things make cloud
analytics fast. All three
now run on your servers.

Cloud platforms are quick because they own the machines and can build for one job. That is also why a regulated team cannot use them. The speed is bought with custody of the data. The 26.01 engine takes the same three mechanisms and runs them wherever you deploy Countly.

  1. 01

    Streaming ingestion

    Events are validated, transformed and written as they arrive instead of collected into batch windows, so a dashboard is current with the product rather than with the last run.

    On your write path
  2. 02

    Analytical storage

    Events and profiles live in a store built for aggregation across large datasets, not in a transactional database bent into the job. That is where the query speed comes from, and why it needs less hardware than the previous generation.

    In your deployment
  3. 03

    Distributed query processing

    Funnels, retention, cohorts and Drill spread across nodes, so exploration stays interactive as the dataset grows instead of degrading into overnight reports.

    On your hardware
GDPRCompliant by design
HIPAASupported deployment model
SOC 2Type II examined
TISAX®Participant*
CCPASupported

*Countly Ltd. is a TISAX® Participant. TISAX® is a registered trademark of and governed by ENX Association. Learn more at enx.com/tisax

By team

One deployment,
four sets of answers

The same event store, read from each side of the organization. Nothing is exported between them.

Product

Funnel analysis, retention, user journeys and feature adoption, computed on raw events rather than on rollups.

Security and compliance

Self-hosted or private cloud deployment, data residency where it is required, and an architecture built to survive the review.

Marketing

Segmentation, personalization, campaign measurement and customer lifecycle analytics, triggered from the same live data.

AI

Real-time user context for assistants and agents, personalization models and predictive workflows, through the API and MCP.

Capabilities

Six parts,
one deployment

Every part runs inside the same environment, on Flex operated by Countly, on Enterprise operated by you. The boundaries are internal, so the data never crosses a vendor line.

SDKs and write API

10+ SDKs for mobile, web, desktop and embedded devices, plus a REST write API for backend systems. Events queue offline and arrive when the device does.

Ingestion pipeline

Event streaming on the write path, where validation, transformation, consent checks and masking run before anything is written, at the volumes a busy product produces. Bad data is fixed or dropped on arrival.

Data control layer

The governed schema for events and properties, access rules per app and role, PII policies and the audit log. Product, engineering and security teams share it.

Event store and profiles

Raw events and merged user profiles in one analytical storage layer inside your deployment, built for aggregation across large datasets rather than for transactions. No pre-aggregation, unlimited retention; sampling only if you turn it on.

Real-time query engine

Distributed query processing, so funnels, retention, cohorts, Drill and dashboards compute on the live store within seconds of arrival, and interactive exploration stays interactive as the dataset grows. The same queries feed reports, campaigns and Cee.

Engagement and AI runtime

Journeys, push, in-app, experiments and Adaptive Variables are triggered from the same store. Cee and your own models read the same live context through the API and MCP.

How it works

How data moves
through Countly

Four stages, one system. Each stage is configurable, none of them leaves your deployment.

  1. 01

    Collect

    SDKs and the write API deliver events and profile updates, online or queued offline.

  2. 02

    Control

    Validation, transformation, consent and masking rules decide what is stored and how.

  3. 03

    Store

    Raw events and merged profiles land in one storage layer, in your region or your infrastructure.

  4. 04

    Serve

    Reports, journeys, experiments and Cee read the live store, within seconds of arrival.

In depth

How it works,
in detail

The parts that matter when you evaluate this area seriously: what it does, how it behaves at the edges, and what your team controls.

Data flow

One path from device to decision

An event leaves the SDK, passes validation and consent rules, is transformed and masked according to the data control layer, and lands in the event store together with the user profile it belongs to. From that moment it is part of every report, every cohort, every campaign audience and every answer Cee gives. There is no ETL and no nightly job between the user and your view of them.

  • Consent and masking before storage, never after
  • Profiles merged on identification, backend attributes via API
  • Same store for analytics, engagement and AI

Topologies

From one host to a multi-region cluster

A Flex instance is a dedicated, single-tenant deployment in the region you choose. Enterprise runs the same system on your infrastructure: a single node for a pilot, a clustered deployment for production, several regions where residency demands it, or an air-gapped network with no outbound connection at all.

  • Dedicated single-tenant instances only
  • Your cloud, your data center or air-gapped
  • Multi-region where residency requires it

AI

Your AI, your infrastructure, your data

Behavioral data is the context an AI needs to be useful about your product, and it is the last data a regulated team wants to send out. Countly keeps it inside the deployment and exposes it to agents and models through the API and MCP, so assistants, personalization models and predictive workflows read live product context without the data leaving your network.

  • Real-time user context for assistants and agents
  • Personalization and predictive models on live behavior
  • Agents connected through MCP, inside the deployment

Scale and isolation

Built for billions of events

Ingestion, storage and query scale horizontally and independently. Production deployments serve up to 100 million monthly active users and billions of events on one system, with isolation that comes from the deployment itself rather than by a tenant ID in a shared database.

  • Horizontal scaling per component
  • Up to 100 million monthly active users per deployment
  • Isolation by environment, not by tenant flag

FAQ

Questions
teams ask us

Straight answers on how this part of Countly works in practice. For anything else, talk to us.

What is new in the 26.01 architecture?

A redesigned analytics engine: event streaming on the ingestion path, high-performance analytical storage instead of a transactional database, and distributed query processing. The result is cloud-grade query speed on materially lower infrastructure than the previous generation needed.

Which deployment options are supported?

Self-hosted on your own infrastructure, or private cloud in the region you choose, with data residency support in both. Customer data stays inside your environment and under your jurisdiction either way.

What does Countly run on?

Standard Linux hosts or Kubernetes, in any cloud or data center. Flex deployments are provisioned and operated by Countly; Enterprise deployments are installed from packages in your environment. Sizing guidance and reference architectures are in the documentation.

Does it scale horizontally?

Yes. Ingestion, storage and query scale independently, so a growing event volume or a new region is a capacity change, not a re-architecture.

How do updates work on Enterprise?

Updates ship as versioned packages you install on your own schedule, including in air-gapped environments. Nothing requires an inbound connection from Countly.

Where does the AI run?

Cee works on the live event and profile store. On Enterprise, Cee’s product agents can run inside your deployment when configured with your own model endpoint. The Documentation Agent and the Countly AI Gateway (EU) run outside.

Architecture

One system.
Your infrastructure.

Talk to an engineerRead the docs