Sits on top of your existing platform. Analyses member behaviour, sends outreach, follows up when ignored, learns what works, adapts. You set the guardrails; it handles the volume.
Five layers, cleanly separated. Only the bottom layer is platform-specific; everything above it works across any community tool.
Send, wait, check outcome, follow up with a different angle or mark dormant. Up to 3 attempts per member. Durable workflows that survive restarts.
The agent reads its own performance data. "Event invitations get 35% opens for tech members vs 5% for check-ins." Adapts approach automatically or proposes changes for approval.
Agent assesses community state, selects relevant skills. New skills are just markdown files. Re-engagement, renewals, onboarding, event promotion, trends, anniversaries.
Persistent observations about members and the community. CM corrections stick across runs. "She's on maternity leave until June." The agent remembers what you told it.
Goals, audience description, never-do list, autonomy level. The purpose layer that makes full delegation viable. First run builds it from conversation.
Per-action gates, not just global auto-send. Auto-send routine emails, but always ask before marking someone dormant. Three levels: auto, propose, approve.
Re-engagement, membership renewals, new member onboarding, event promotion, engagement trends, community health. Each with its own rules and autonomy settings.
Every decision logged with reasoning: what the agent chose, why, what data it used, who approved. GDPR and EU AI Act compliant.
Customer success teams manage dozens of accounts. The same agent that runs one community can monitor a portfolio -- surfacing risk, preparing talking points, and keeping every account warm without context-switching.
One summary across all your accounts: who needs attention today, which renewals are approaching, where engagement is trending down. Start the day with a plan, not a spreadsheet.
Engagement, renewal rates, and activity trends tracked per account and across the portfolio. Spot patterns that individual dashboards miss.
Flags accounts showing early warning signs -- dropping logins, ignored emails, lapsed event attendance -- before the renewal conversation starts.
Generates quarterly business review materials from real activity data: engagement highlights, risk areas, recommended actions. Ready to present, not just raw numbers.
Trust is the barrier, not technology. Each phase earns the right to the next.
Multi-step follow-ups via Temporal. Outcome-driven learning that adapts strategy per segment. Skill discovery from markdown files. Persistent agent memory. Community brief. Per-action autonomy gates. Six workflows, full audit trail.
Combine re-engagement + renewal into one message. New skills: onboarding journeys, member matching, moderation, board reporting. New platform actions: event CRUD, profile updates. New community onboarding flow.
The agent runs programmes, not just tasks. Multi-platform: connect Hivebrite, Circle, Discourse, or any community tool. Agent proposes new skills. Weekly briefing replaces daily queue.
The architecture separates cleanly: workflows and dashboard are generic, only the platform connector is specific. Adding a new platform means writing one connector -- not rebuilding the product.
Today it connects to Hivebrite. The same approach works for Circle, Discourse, Mighty Networks, or any platform with a member API. One agent layer, many platforms.