meeting recorders aren't note-takers. they're a firehose.

i use circleback as the input layer for my entire business ops agent. his name is mo. every meeting i take feeds a system that classifies, processes, routes, and compounds context automatically.

most recorders could technically do this. circleback's system is dead simple and i like to keep things that way.

here's the full pipeline.

1. push. all meeting notes land in a central vault immediately after each call via webhook. separate repo, separate from code. intentional separation between the codebase and the content it operates on.

2. classify. meetings get tagged by title. simple regex.. internal, external, 1-1, standup. nothing fancy.

3. process. each meeting type has its own schema. structured data gets extracted based on classification, then routed to the right slack channel.. development, biz-dev, product, or a DM. every output validates against a schema. no hallucinated structure.

4a. update entities. if the meeting references known proper nouns.. companies, products, concepts, competitors.. mo adds new touchpoints to existing knowledge graph records. tracks sentiment shifts, relationship changes, strategic context. institutional memory that compounds.

4b. update contacts. if external people were on the call, mo enriches lead and prospect records with conversation context. tracks relationship strength across all touchpoints. feeds the CRM without anyone typing anything.

4c. update product specs. if features or requirements came up, mo adds citations linking specs back to the source meeting. flags when discussed requirements drift from existing specs. keeps product truth grounded in actual stakeholder quotes.

4d. update action items. every task links to its origin meeting. full citations.. who said it, when, in what context. no orphaned tasks. everything traces back to a real conversation.

5. reconcile gaps. a daily cron checks my calendar for attended events missing notes. gaps found? triggers circleback MCP backfill. fetches the meeting, runs it through steps 2 through 4.

6. query anything. via slack or CLI, mo can pull from the full vault.. circleback MCP plus the processed meeting notes table. search across all historical meetings by topic, person, or timeframe. cross-reference what was said against what we built. answer questions like "what did we promise X company?" or "when did we first discuss Y feature?"

7. weekly synthesis. automated rollups generated from real conversations. not status updates. priorities extracted from what people actually talked about. decision history with full attribution.

the philosophy behind all of this.. meetings are where the actual work happens. unstructured conversations, when properly processed, represent the ground truth of what you're building and why.

not the sanitized status update. not the aspirational roadmap. the real discussions, commitments, and context.

like katamari.. you roll through your day and everything sticks. the ball just keeps getting bigger. except this one is queryable.