The Adjutant
The AI that won't let your business forget what it learned.
Most dashboards show you what happened last week. You look at the numbers, maybe notice something, maybe don't, and move on. The insight dies in your head.
The Adjutant closes those loops. When your sales team takes calls, it pulls the transcripts, extracts the patterns — which objections keep coming up, which closers are improving, where the script is failing — and posts a summary to your sales channel. Your sales manager reviews it, adds corrections, and those corrections feed back into next week's analysis. The system learns.
It does this across sales, operations, and strategic planning. If a metric is missing, a meeting had no recording, or a decision has been pending for two weeks, it flags the gap and asks about it. It doesn't wait for you to notice — it prompts you. That's the difference between a display system and a learning system.
Seven cron jobs form the closed-loop architecture, each following a five-step template: define outcome → produce output → measure against outcome → write delta to semantic memory (nova_memory) → next iteration reads the delta.
The sales loop runs weekly: call-analysis.ts aggregates Fathom transcripts, pattern-extraction.ts identifies recurring objections and skill gaps, results post to #sales-loop-ai in Slack. Brandon (sales manager) reviews in-thread — corrections feed back via the next extraction run. The compass loop pulls git history, Slack digests, cron health, stale metrics, overdue tasks, and meetings without recordings into a gap detection pipeline. Recommendations auto-create tasks; unresolved gaps route to SuperNova (Telegram AI) for follow-up.
All outputs write to nova_memory with governance columns (scope, visibility, sensitivity, TTL, recurrence count). Pathfinder — the semantic retrieval gateway — reads from this memory with ranked precedence and domain filtering, so every future AI session (Claude Code, Nova, briefings) has access to what the system learned, not just what it displayed.