Value-led prospecting & advice copilot
About this agent
A growth assistant for insurance brokers, built to earn a prospect's trust with real help before any product conversation. A raw inquiry — a policy question, a claim stuck mid-hospitalisation — opens a session in which callLLM answers directly, cross-references market benchmarks pulled read-only from a connected policy database, and quietly extracts the underlying health, employment and coverage signals as the memory is saved. Closing that session promotes what was learned into a shared PROSPECT_PIPELINE project, where it stays retrievable weeks later when an advisor drafts a handoff message or a lead-intelligence card, without needing the original conversation to still be open — sessions are single-use once closed. A policy audit benchmarks a prospect's own policy against market standards and flags gaps such as co-pay percentages or sub-limits in plain language, without naming or pushing a competitor's product. A separate command generates short, reusable educational content for a given audience and topic, checked against a company-wide cache first so a near-duplicate isn't regenerated and billed for twice. Multi-step lead qualification — a parsed document note, a pipeline-stage update, a follow-up reminder — is staged as one zenmem transaction so a partial write never lands. It never pressures a prospect toward a purchase; the value comes first.
What changed with Zenmem?
The same agent, built twice against the same contract — once on Zenmem, once on MongoDB + LangChain/LangGraph.
Before → after
Before With Zenmem
What the team gained
- A handoff draft built weeks after the conversation reads purely from project scope, so the original session never has to be reopened or replayed.
- Session scope holds the live prospect conversation and promotes only the extracted context on close, keeping the pipeline free of raw transcript.
- Niche educational content is company-scoped and checked before regeneration, so the same explainer is not paid for twice.
- The lead-intelligence card is a read over what was promoted, not a record that has to be kept in step with the conversation.
- A connected-database fetch brings policy facts into the same call, with no export job between the two.
How memory is scoped
`scope="session"` holds one prospect's raw conversation — the inbound inquiry, a policy audit, moment-of-need guidance — for as long as it's active; ending it promotes what was extracted into long-term storage. `scope="project"`, fixed to PROSPECT_PIPELINE, is where that promoted context lives afterward: weeks later, an advisor's handoff draft and lead-intelligence card are built purely from this project-scope memory, without reopening the original session. `scope="company"` holds niche educational content, reusable across every prospect rather than tied to one lead, and checked before regenerating it. A connected read against DB-BENCHMARKS / DB-POLICIES supplies market benchmarks for the policy audit and moment-of-need answers, and is never written back to.
How it works
From a raw inquiry to a warm advisor handoff.
Answer the moment of need
A prospect's raw inquiry or claim problem gets a direct, non-salesy answer grounded in connected-database market benchmarks, inside a fresh session.
Extract the signals
Health, employment and policy-frustration signals are quietly saved to session memory as part of that same call.
Promote to the pipeline
Ending the session flushes everything into the shared PROSPECT_PIPELINE project, retrievable long after the conversation ends.
Draft the handoff
Weeks later, an advisor pulls the accumulated project-scope context to draft a warm, non-pushy consultation offer and a lead-intelligence card.
Nurture the niche
Targeted educational content per audience and topic is generated once, cached at company scope, and reused rather than regenerated.
Commands
Every capability runnable in one line.
Commands
- agent7 ping — liveness check against the configured zenmem deployment.
- agent7 audit <policy_details> — unbiased policy gap analysis benchmarked against DB-BENCHMARKS / DB-POLICIES.
- agent7 claim-help <situation> — moment-of-need claim or admin guidance and document checklists.
- agent7 inquire <user_query> — opens an inbound lead session, answers directly, and extracts prospect signals into session memory.
- agent7 handoff <query> — drafts an advisor handoff message and a lead-intelligence card from project-scope memory.
- agent7 nurture --audience <a> --topic <t> — generates a niche educational content snippet, cached at company scope.