Signals
Talk to an engineerStart freeSnowplow
Product/Agentic context

Your agent already knows what the customer just did.

No more "how can I help you today?" to a customer who has been stuck on the same page for four minutes. Agentic context hands your agent their recent activity as JSON, or as a plain-text block ready to drop into a prompt, current as of this session.

Start freeTalk to an engineer

The agent below is reading your session. It will speak up to five times on its own, then wait for you.

Signals returns this

connecting

Waiting for the first read of your session.

t+seventevent_context
waiting for your first event…
The same JSON your agent receives, by API request or through the MCP server.

Your agent uses it

The agent is reading your session. It will speak first, on its own.

The problem

The agent answers well. It cannot see what the customer just did.

The agent sees the conversation, and at best a CRM record that says who the customer is. What it does not see is that they have been stuck on the checkout page for four minutes and searched twice for the returns policy. So it opens with "how can I help?" and the customer has to explain what your own site already knows.

The same customer message in an OUTPOST support chat answered twice: with no context the agent asks how it can help, and with agentic context it answers about returns and the cart waiting at checkout.
In Snowplow Console

Choose the events, the window and the prompt.

An agentic context keeps the recent events you pick, up to a count and an age you set, with the instruction your agent reads alongside them.

The Snowplow Console showing an agentic context with its event window, prompt and the event properties it includes.
Snowplow Console, shown with example data.
How it works

The same events. Shaped for a model. Read at the moment the agent needs it.

Three tracked events from a shopper's session: add to cart, a checkout page view and a search for the returns policy.01 · Event in

The same SDKs and the same events that power attributes and triggers.

Nothing extra to instrument. Schema-validated on the way in.

An agentic context definition that keeps the last 20 events or 600 seconds of activity and returns it as JSON or narrative.02 · Shaped

Recent session activity becomes a compact context.

Structured JSON for your code, or a plain-text block, a summary line and a table of events, ready to drop into a prompt. Same source, your choice of shape.

An agent reading the context with getAgenticContext, available from Python, Node.js, the REST API and the MCP server.03 · Read

The agent asks mid-conversation and answers with what it now knows.

Python or Node.js SDK, REST API, or the MCP server. Pair it with live attributes so the agent knows both the pattern and the moment.

Works with
  • Claude
  • LangChain
  • LangGraph
  • LlamaIndex
  • CrewAI
  • OpenAI Agents SDK
  • Vercel AI SDK
  • Mastra
  • Pydantic AI
  • Google ADK
  • CopilotKit
  • Amazon Bedrock
  • Vertex AI
Capabilities

Recent activity, shaped so a model can use it.

The same recent checkout events read two ways: as JSON with a summary, prompt and events for your code, and as narrative, a plain-text block with a summary line and an event table, for a prompt.
01

JSON for your code, or a plain-text summary and event table for a prompt. Same source, your choice of shape.

The same recent checkout events read two ways: as JSON with a summary, prompt and events for your code, and as narrative, a plain-text block with a summary line and an event table, for a prompt.
02

Built from the same event stream as attributes and triggers, so it reflects the last few minutes, not last night.

An agentic context listing this session's events from the last few minutes, next to a nightly warehouse snapshot loaded 12 hours ago that does not include today's session.
03

Combine recent activity with computed attributes so the agent knows both the pattern and the moment.

Recent events from getAgenticContext and attributes from getServiceAttributes, such as lifetime value and loyalty tier, combined into what the agent reads.
04

Python and Node.js SDKs, the REST API, or the MCP server. Ask for a customer's context as one tool call.

An OUTPOST support chat where the agent calls getAgenticContext in narrative format, gets a summary line and a table of the session's events, and replies with the returns policy and help with the $184.20 cart.
05

Each context follows one session and keeps the most recent events, up to the count and age you choose. Older activity drops out on its own.

One session's events on a timeline, with a window that keeps at most 20 events no older than 600 seconds, older events falling out, and a second session with its own window.
06

Customer activity never leaves your deployment on the way to the model.

Your cloud containing the event pipeline, Signals and your backend, with your backend building the prompt and sending only that prompt to the model provider you choose.

Full reference: Defining agentic contexts →

Demo

The same question, asked twice.

A shopper compares two jackets, opens a size guide, then asks the agent which one to pick. Once without agentic context, once with it.

Recorded in a demo store. The panel on the left is the event stream the agent reads.1 min 44 sec

Collect. Define. Serve.

Add the SDK, define one attribute, and watch it change while you click around your own product. The free tier runs on the real engine: no card, no sales call.

Start freeTalk to an engineer