Behavioral memory for your agent.
Agentic context hands your agent what the customer did this session, as JSON or a plain-text block for the prompt.
The agent below is reading your session.
An agent is talking to you
connectingWhat the customer just did reaches your agent too late.
The agent sees the conversation and whatever profile it is handed. What the customer did on the site this session, such as four minutes stuck on checkout and two searches for the returns policy, arrives after the chat has started. So it opens with "how can I help?" and the customer has to explain what they just did.
Signals shapes session events for your agent to read mid-conversation.
We host it, or it runs in your own AWS or GCP account. Signals collects events with its own SDKs, updates what it knows within a second of each event, and answers a read in 6ms at the median and 10ms at p95, in-region.
Events come from the Snowplow SDKs that also feed attributes and triggers.
Add the SDK once and all three read the same events, schema-validated on the way in.
Recent session activity becomes a compact context.
JSON for your code, or a plain-text block with a summary line and an event table, ready to drop into a prompt.
The agent asks for it as a tool call and answers with what it now knows.
Python or Node.js SDK, REST API, or the MCP server.
- Claude
- LangChain
- LangGraph
- LlamaIndex
- CrewAI
- OpenAI Agents SDK
- Vercel AI SDK
- Mastra
- Pydantic AI
- Google ADK
- CopilotKit
- Amazon Bedrock
- Vertex AI
Recent activity, shaped so a model can use it.
At checkout with a $184.20 cart, after three searches for the returns policy.
Gold tier, $1,240 lifetime value, three failed searches this session.
Returns are free within 30 days, so you are covered. Your $184.20 cart is saved at checkout.
JSON for your code, or a plain-text summary and event table for a prompt. Both come from the same events.
Built from the same event stream as attributes and triggers, so it includes what the customer did in the last few minutes of this session.
Read it alongside computed attributes, such as lifetime value or loyalty tier, so the agent has the customer's history and this session together.
At checkout with a $184.20 cart, after three searches for the returns policy.
Gold tier, $1,240 lifetime value, three failed searches this session.
Returns are free within 30 days, so you are covered. Your $184.20 cart is saved at checkout.
In a private deployment, Signals and your backend run in your cloud. Your backend decides what goes into the prompt it sends to the model provider you choose.
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.
Agents and pages that start from this session.
session context
- 2 jackets viewed
- size guide opened
- pickup slots error
As text for a prompt, or JSON for code.
Frequently asked questions.
We already pass a CRM profile to our agent. Why add this?
Keep the profile. It says who the customer is and what they have bought. Agentic context adds what they did in this session, such as the pages they viewed, the searches they ran and the errors they hit, so the agent can answer about the checkout they are stuck on. Your backend puts both into the prompt.
Does customer data go to the model provider?
Only what your backend puts in the prompt. You choose which events and properties an agentic context includes, Signals serves it to your backend, and your backend builds the prompt and sends it to the model provider you choose. In a private deployment, Signals runs in your own AWS or GCP account beside your backend.
Which agent frameworks does it work with?
Any framework that can call the REST API or the MCP server, including LangChain, LangGraph, LlamaIndex, CrewAI, the OpenAI Agents SDK, the Vercel AI SDK, Mastra, Pydantic AI, Google ADK and CopilotKit. Python and Node.js SDKs are there too. None of these is a packaged integration. Reading a customer's context is one tool call.
Our agent framework has memory. How is this different?
Agent memory, in your framework or in tools such as mem0, Zep and Letta, recalls what the customer said in the conversation. Agentic context holds what the customer did on the site this session, such as pages, searches, errors and time on a step. They sit side by side, one for recall and one for behavior.
How far back does it go?
An agentic context covers one session, up to the number of events and the age you set, and older events drop out on their own. For what a customer did in earlier visits, read attributes keyed on the user beside it.
Start with one agentic context.
Add the SDK, choose the events your agent should see and the window, and hand it the session as one tool call, in JSON or narrative. The free tier runs on the real engine: no card, no sales call.