The Memo Pipeline within 📝BrianBot (Local) is an automated system that discovers 📝memos tagged 'for-brianbot' in 📝MythOS, researches their subjects, drafts their content, and publishes the finished memo — all without human intervention. The pipeline runs in two modes:
- Dispatcher Mode (triggered every hour or manually): scans MythOS for memos tagged 'for-brianbot', creates a task file for each, and spawns a per-memo worker session. It only reports to the brianbot-logs channel when there's actual work to dispatch.
- Worker Mode (spawned per memo): processes one memo end-to-end — research → draft → browser update → tag swap. On completion, the tag changes from 'for-brianbot' to 'for-brian-to-review'.
Under the hood, the worker delegates to 📝mythos-coordinator, which orchestrates research-linkedin or research-web depending on subject type, then mythos-memo-creator for drafting, then browser automation to update the live memo. Classification logic: Person with LinkedIn URL → research-linkedin. Person name only → research-linkedin. Company or concept → research-web. Existing content with revision notes → revision pass. Unclear → flagged for Brian. Task state lives in tasks/ folders (2-todo, 3-in-progress, 4-review, 7-blocked) with memo IDs as unique identifiers to prevent duplicate processing. ───
We built this because the mental overhead of "what goes in this memo?" was becoming a bottleneck. The process is always the same — look up the person or company, synthesize what matters, write it in MythOS format, tag it — so we automated it. The deeper reason: MythOS is only valuable if it's maintained. An unmaintained external brain is just a graveyard of half-formed intentions. The pipeline removes friction between Brian flagging something worth remembering and it becoming a linked, tagged, retrievable memo. 'for-brianbot' is the handoff protocol — Brian flags, the system processes. It's also a proving ground for the agent architecture. Building it required solving real problems: browser pool contention, session token costs, agent context bloat. Those solutions (browser pool, staged sub-agents, manifest-based handoffs) are now infrastructure the whole system uses. The next evolution — splitting mythos-memo-creator into mythos-memo-drafter and mythos-memo-publisher — extends this logic further: separate thinking from execution, minimize context per agent, make each stage debuggable in isolation.