Give the agent the last five minutes.
An agent that opens with yesterday's profile is guessing. Signals hands it this customer's recent activity in a form a model can use, JSON or narrative, over an API or MCP, before the first reply.
- 00:00Customer opens the chat after four minutes stuck on one page
The ticket shows last month's profile.
- 00:01
recent_activityrequested for this sessionRecent activity as JSON or narrative.
- +6msThe agent reads it before the first reply
It already knows which page, how long, and what failed.
The agent has amnesia.
Raw events are unreadable to a model, and the profile in the CRM was written overnight. So the agent asks the customer to describe what just happened, when the events describing it already exist.
One call returns what the customer has been doing, in words or JSON.
Choose the activity the agent should know about
Pages viewed, searches, errors hit, time on step. Live attributes plus the recent event log.
Pick JSON for tools, narrative for prompts
Agentic context is the same activity rendered either way, so the model reads it without a parser.
Call it from your agent framework or over MCP
Works with the agent you already run. Signals supplies the context; the model decides.
| Attribute | Key | Window | Read by |
|---|---|---|---|
minutes_on_current_page | session | live | Agent system prompt |
errors_this_session | session | this session | Escalation rule |
recent_activity | session | last 30 events | Agent context window |
The first reply knows what happened.
Before the first reply, p50 in-region
JSON for tools, narrative for prompts
Shared by the agent and the site it is helping with
user · session · basket · listing · store
Start with one attribute.
Define it, read it in your own product, and see it change while you click around. 14 days, no card, no sales call.