For AI agents: a Markdown version of this page is at https://docs.snowplow.io/docs/signals/introduction.md. The index of all documentation pages is at https://docs.snowplow.io/llms.txt, and the index for this section is at https://docs.snowplow.io/docs/signals/llms.txt.
Snowplow Signals overview
Snowplow Signals computes user attributes from your behavioral data in real time, and serves them to your applications over an API. Attributes are calculated from your Snowplow event stream as events arrive, or synced from tables in Snowflake, BigQuery, or Databricks.
Use Signals to:
- Personalize content, recommendations, and pricing based on what the user is doing in their current session
- Serve scores and predictions calculated in your warehouse, such as churn propensity, to your applications through one API
- Give chatbots and AI agents context about the user they're interacting with
- Trigger actions automatically when users meet criteria you define, such as offering a discount when a high-value cart is abandoned
- Serve up-to-date behavioral features to ML models, for use cases like lead scoring
New to Snowplow: the Signals free tier is a hosted environment, with event collection included, capped at 5 million events a month, with no end date or credit card. Snowplow runs it on AWS, in the US or the EU depending on the region you sign up from.
Already run Snowplow CDI: Signals reads events from your existing pipeline, and is deployed in the AWS region you choose. See Set up Signals.
To try Signals on your own web app, follow the quickstart, where a coding agent adds the tracking and configuration for you, or the step-by-step quickstart to do the same steps in Console.
Why teams use Signals
Signals handles collection, calculation, and serving:
- Collection is built in: your Snowplow pipeline tracks and validates events, so you don't build a separate event pipeline, stream processor, or online store.
- Attributes are short definitions, not code: choose the events, an aggregation, and a time period, such as "product pages viewed in the last 10 minutes".
- Values update in the session: attributes update as each event arrives, so your application reads current values while the session is still in progress.
- One read combines live and historical data: a service returns stream attributes alongside warehouse-synced values.
- One definition serves real time and training: the same attributes build point-in-time datasets for ML models.
- Your systems decide: Signals serves the values, and your application, model, agent, or personalization engine decides what to do with them.
Why teams choose Signals over other tools
| Instead of | Why teams choose Signals |
|---|---|
| A feature store | A feature store serves data you already collect and process. Signals collects and processes the events too, so you don't build ingestion. |
| A customer data platform (CDP) | CDPs update profiles and audiences in batches for marketing tools. Signals updates attributes as each event arrives, for use in the current session. |
| Building it in-house | You don't build, run, or maintain your own collection, stream processing, low-latency store, or API, so your engineers can work on your product instead. |
Many teams keep their CDP or feature store alongside Signals. For named vendor comparisons, see Evaluate Signals.
How Signals works
Signals calculates attributes from your events and stores the latest values in the Profiles Store, where your applications and agents read them.
- A Snowplow tracker sends events to your pipeline.
- Signals updates the values of your attribute groups as each event arrives, or syncs them from warehouse tables.
- Applications read the values through a service. Agents read the user's recent events through an agentic context, or attribute values through the Signals MCP server.
- When values meet an intervention's criteria, Signals pushes the intervention to subscribed applications.
The components you configure:
- Attribute group: a versioned set of attributes, such as
product_view_count_last_10_min, calculated for one attribute key. - Attribute key: who or what attributes describe, such as
domain_useridfor a user ordomain_sessionidfor a session. - Service: attribute groups pinned to specific versions, for applications to read.
- Intervention: an if-then rule on attribute values. The first time a user meets the rule, Signals pushes the intervention to subscribed applications, such as the browser plugin.
- Agentic context: the user's most recent events in the current session, up to 100 events and no older than 1 hour, as JSON or a narrative for an agent.
See How Signals works for detail on each.
Find the right page
| I want to | Page |
|---|---|
| Try Signals on my own web app | Quickstart |
| Enable Signals on an existing Snowplow CDI account | Set up Signals |
| Send events for Signals to calculate from | Send events to Signals |
| Calculate attributes from events or warehouse tables | Define attributes |
| Read attributes in my application | Services, then retrieve attributes |
| Ground an AI agent in the user's current session | Define agentic contexts |
| Let an agent read Signals data as tools | Signals MCP server |
| React when a user meets criteria | Define interventions, then subscribe to them |
| Build training data for an ML model | Create ML training datasets |
To manage Signals from a chat instead, connect your assistant to the Snowplow MCP server, or use the Snowplow Assistant in Console.
Try a tutorial
Each tutorial builds a working demo on top of Signals. See all tutorials, or visit the Developer Hub for videos, blog posts, and workshops.