Insights
Save an objective across several agents and get ranked, cited findings of what is worth doing now, before anyone has to ask.
Switched on per account
Insights are off until they are switched on for your account. Until then the dashboard hides them and the MCP insight tools answer not_found.
Most teams only find out what matters when someone thinks to look: a support lead opens tickets, a recruiter scans the news, a lawyer checks a renewal date. By then the customer has already called, the vendor has already briefed another agency, or the notice window has closed.
Insights reverse that. You write the standing question once. Hyperstruck asks the agents that already know your customers, contracts and market, and puts the few things worth doing today on a feed, each one cited. The time you save is the time nobody spent hunting, reconciling and deciding what to do first.
What you get back
Each finding is one subject, one recommended action, and a reason it cannot wait:
| Part | What it tells you |
|---|---|
| Priority | high, medium or low, so the feed leads with what matters most |
| Why now | The dated facts that make this week different from last week |
| Recommended action | Who to contact, what to offer or what to change |
| Draft message | A ready-to-send note, specific to the evidence, with no promises the records do not support |
| Evidence | The claims, learnings and obligations it rests on, each pointing at the agent that holds them |
A finding that cannot cite a fact about its subject is never shown. What it leaves off the feed matters as much as what it puts on: a customer with one fixed outage, an account that already renewed, a company in a hiring freeze.
Here is the feed after Candidate rediscovery has run for a recruiting desk. Three findings are open, from eight insight rules.
The first card is high priority: Amelia Chen for Juniper Labs' Founding CTO role, 96% supported, from Candidate rediscovery. The card cites claims from the client-desk agent.
The list leads with a person and a role, not a topic. A banner on this card notes that part of the evidence is still waiting on a decision, so the draft is held until someone settles it. The claims underneath come from the agents that already held the brief and the conversation, not from a search you ran by hand.
Before you start
An insight is only as good as the agents it asks. Seed those agents first:
- Documents so they have the notes, tickets, contracts and records your team already writes.
- Claims so those records become dated, sourced facts about people, companies and things.
- Learnings so each agent also holds the playbook: what to do when this kind of situation comes up.
Claims search is how an insight asks an agent a factual question. The learning loop is how everyday work keeps those records current. An insight over empty agents returns nothing useful; that is expected, not a fault in the insight.
When to use an insight
Asking one agent a question is still the right move for a one-off. An insight is the right move when the question should keep being asked, across more than one specialist, without someone remembering to do it.
| You want | Use |
|---|---|
| An answer to one question, right now, from one specialist | Ask that agent directly, or search its claims |
| A standing watch: "tell me who to call this week, before they call us" | An insight |
| Facts split across teams (support and network, market signals and the GTM desk, contracts and policy) | An insight over those agents together |
| A ranked list with "why act now", so a person spends their morning on the few that matter | An insight, then the feed |
Picture it in your own work: the recruiting desk that should hear about a US vendor setting up in Australia in time to offer the first hires; the account team that should follow up an expansion signal before the budget locks; the counsel who should see a renewal that breaks policy while there is still time to give notice. Those are insights. The examples walk two of them end to end.
How it compounds
Insights do not replace the intelligence you already invested in. They spend it.
Every document you send, every run you observe, every claim a person corrects, makes the next insight run sharper. Findings cite those records, so a better claim or a proven learning shows up as a better "why now" and a more specific draft. When you mark a finding actioned or dismissed, the next run does not repeat the same facts. A genuinely new fact about that customer can still surface. The loop is the same as the rest of Hyperstruck: the work you already do makes the next look cheaper.