Two-sided learning
Search insights (which leads convert) AND contact insights (which phrasings, cadences and channels actually book calls). Each side is bounded at ~1500 chars · the playbook never bloats.
Most AI sales tools fire the same prompt forever. CloseHunt agents update their own playbook. Every ~6 hours, once an agent has accumulated 10+ outcomes (replies, conversions, terminations), the engine rewrites a bounded summary of what's working and what isn't, and injects it into every subsequent prompt. Per-agent only. Never shared. Always explainable.
Search insights (which leads convert) AND contact insights (which phrasings, cadences and channels actually book calls). Each side is bounded at ~1500 chars · the playbook never bloats.
Refresh fires max once per side per 6h AND only after ≥10 new outcomes. No early refreshes on thin data, no over-refreshing on volume spikes.
Uses Haiku or GPT-5 mini for the refresh pass. Each cycle costs fractions of a cent. The learning loop runs forever without inflating your AI bill.
Insights live on the agent row, isolated from every other agent in the workspace. Your support agent doesn't learn from your outbound agent. Clean separation.
Clone an agent and you clone its learned playbook. Spin up a sibling agent in a new geo or vertical without losing three months of accumulated wisdom.
Every agent surfaces a 'What this agent has learned' panel · winners, losers, last refresh date, sample counts. No black box · you audit what the AI thinks works.
Fine-tuning rewires model weights and takes hours and dollars per update. Adaptive memory is a few hundred chars of context injected into every prompt · refreshable in seconds, costing fractions of a cent. Same convergent effect, zero MLOps.
Yes. The refresh isn't append-only · every 6h the engine rewrites the WHOLE summary based on the latest 10+ outcomes. Bad lessons drop off the bottom as new evidence accumulates.
Replies (categorized as positive, negative or neutral), conversions (goal-reached), terminations (opt-outs, hard-no, max-turns reached). Each outcome bumps the sample counter; the threshold gates the next refresh.
No. Adaptive memory runs strictly inside your workspace. Insights are stored on your Agent row, never leave your tenant, never share with another workspace, and never get added to model training data.
Yes · that's the whole point. Old insights age out as the engine rewrites them from the latest evidence. A pricing change, a competitor launch, a new vertical · the agent's playbook updates within hours, not weeks.
Trial on request · contact our sales team.