Snowplow

Validated, enriched event data is delivered into the customer's warehouse, lake or stream, events are forwarded to other tools, and Signals context is consumed by agent frameworks through its SDKs and API. These move data to systems and agents; no connectors that let an agent take authenticated actions in outside systems are documented.

SourceSnowplow, snowplow.io llms.txt and homepage, and docs.snowplow.io signals/conceptsread 2026-09-21

Data moves through a fixed collect, enrich and deliver pipeline, and Signals adds versioned attribute groups, services that pin versions for applications, and if-then intervention rules with multiple conditions. These sequence Snowplow's own data processing and fire rules; no workflows that mix deterministic nodes with agent steps, or first-class branching, retries and fallbacks for an agent's work, are documented.

SourceSnowplow, docs.snowplow.io signals/conceptsread 2026-09-21

Signals can sync any pre-calculated warehouse tables, such as transactional data or CRM attributes, into its Profiles Store and compute attributes keyed by product, page, campaign or segment, which applications and agents retrieve by key through services. That serves company data to agents, but as lookups of computed values, and no document ingestion, semantic search, visible retrieved passages or source citation is documented.

SourceSnowplow, docs.snowplow.io signals/conceptsread 2026-09-21

Signals resources follow a draft-then-publish model for the people configuring them, and the Snowplow MCP server's assistant acts with the permissions of the user or key that connected it. No approval step, escalation rule or pause for sign-off before an agent's action, whether the Snowplow Assistant's console changes or an intervention a customer's agent receives, is documented.

SourceSnowplow, docs.snowplow.io llms-support/snowplow-mcp and signals/conceptsread 2026-09-21

An ISO 27001 badge sits in Snowplow's site footer alongside GDPR, CCPA and HIPAA marks and a link to a Security page. Console API keys control customer access with permission levels per feature, from View-only to Global admin, and MCP connections act only with the connecting user's or key's permissions.

SourceSnowplow, snowplow.io homepage footer and docs.snowplow.io llms-support/snowplow-mcpread 2026-09-21

Snowplow documents instrumenting an AI chatbot with client-side, server-side and agent self-tracking, so an agent's actions can be captured as events, and the Snowplow Inspector shows Signals attributes and interventions as they fire. That records agent behavior only as far as the customer instruments it; no built-in step-by-step view of an agent's prompts, tool calls and outputs is documented.

SourceSnowplow, docs.snowplow.io llms.txt (agentic self-tracking tutorial, Inspector for Signals)read 2026-09-21

Calculated attributes about each user and session are kept in the Profiles Store in Signals, keyed by user or session identifiers, as period-based, lifetime, first-touch and last-touch values with a configured time-to-live that removes stale data, and agentic contexts capture a rolling window of a user's recent session activity with instructions, returned as JSON or a plain-language narrative for an agent to read. That is memory scoped by user and session, with a stated lifetime, that an agent reads as context.

SourceSnowplow, docs.snowplow.io signals/concepts and snowplow.io llms.txtread 2026-09-21

Snowplow offers three deployment models, Cloud, Private Managed Cloud and Self-Hosted, with Private Managed Cloud running the pipeline inside the customer's own VPC on AWS, GCP or Azure, and Signals infrastructure deployed alongside the pipeline against the customer's warehouse.

SourceSnowplow, snowplow.io llms.txt and FAQ, and docs.snowplow.io signals setupread 2026-09-21

Snowplow publishes solution accelerators for Signals-powered agents (with AWS Bedrock AgentCore, the Vercel AI SDK, Google ADK and a travel-site chatbot) run from a Signals Sandbox, dbt model packs for digital and ecommerce analytics, and an MCP plugin bundling six skills. The agent accelerators are blueprints to build from in a sandbox and the model packs are analytics models, not ready-made agents a buyer adopts.

SourceSnowplow, docs.snowplow.io llms.txt and snowplow.io llms.txtread 2026-09-21

Signals interventions are rule-based triggers that fire automatically when a user's calculated attributes meet their criteria, and are streamed in real time to every subscribed application, with agentic chatbots named as a use case; direct interventions can also be pushed from the API. These events reach the customer's agent from Snowplow's own platform without a person initiating them.

SourceSnowplow, docs.snowplow.io signals/conceptsread 2026-09-21

Signals returns context for whatever model the customer's agent runs, and the Snowplow MCP server works from the customer's own AI tool, but no choice of the model behind Snowplow's own Assistant, and no routing or provider selection within Snowplow, is documented, so for Snowplow's own AI, model choice is not stated as the customer's.

SourceSnowplow, docs.snowplow.io llms-support/snowplow-mcp and signals/conceptsread 2026-09-21

Signals is reached through Python and Node.js SDKs and a REST API with deployment credentials, trackers and SDKs cover web, mobile and server languages, and a remote MCP server authenticated by OAuth or a permission-scoped Console API key gives AI assistants read and write tools over schemas, tracking plans, pipelines, failed events and enrichments, installable with a plugin that bundles six skills.

SourceSnowplow, docs.snowplow.io llms-support/snowplow-mcp and llms.txt (Signals connection)read 2026-09-21

Before an intervention is published, Signals previews show which users it would target, and the Snowplow Inspector debugs attributes and interventions against live tracking, so the rules that wake a customer's agent can be tested before production. No scoring of an agent's output quality over time is documented.

SourceSnowplow, docs.snowplow.io llms.txt (Signals March 2026 release, Inspector for Signals)read 2026-09-21

Snowplow collects and serves behavioral data and context, and no browser, desktop or computer control by an agent is documented.

SourceSnowplow, docs.snowplow.io llms.txtread 2026-09-21