Meeting Preparation

How to generate meeting briefs, save them for later, and use them to walk into every client meeting fully prepared.

Generating a Brief

"Prep my meeting with Acme -- we're discussing Q3 budget and the creative approvals"

Armbrain takes the topics you mention and builds a structured brief by pulling together:

1. Agenda-tagged memories -- anything explicitly flagged for upcoming meetings
2. Topic-driven matches -- semantic search for each topic you mention
3. Meeting-linked tasks -- open tasks and commitments whose source_ref matches the meeting log
4. Recent commitments -- open action items from the current and previous quarter
5. Permanent knowledge -- stakeholder info, preferences, and standing facts
6. Recent backfill -- recent activity that fills remaining context slots

The result is a markdown brief with sections:

• Key Stakeholders
• Action Items
• Recent Decisions
• Active Campaigns
• Key Context
• Stakeholder Notes

Meeting Types

You can specify a meeting type to adjust retrieval emphasis:

"Prep a planning meeting with Acme for Q2 kickoff"

Saving Briefs

By default, briefs are generated and returned but not persisted. To save:

"Prep my meeting with Acme and save it"

This writes the brief to the meeting_briefs table. You can retrieve it later:

"Show me my last brief for Acme"

Saved briefs capture the meeting type, topics, memory count, and the full markdown at the time of generation.

How Many Memories Do I Need?

The brief system works best after 3-4 meetings worth of ingested content. With fewer than 3 memories, Armbrain returns raw data instead of a structured brief (it tells you there isn't enough context yet).

The sweet spot: after consistently pasting transcripts for 2-3 weeks, briefs become remarkably good. The more you feed it, the better it gets.

Saving Briefs to Disk

If you've configured a local folder for the client, you can save the brief as a file:

"Save that brief to Acme's folder"

Armbrain writes the brief into the client's local Briefs folder:

~/Clients/Acme Corp/Armbrain/Briefs/2026-03-20-brief.md

How Retrieval Works

The meeting prep retrieval strategy is designed for balance, not just relevance:

1. Type-balanced -- pulls from decisions, commitments, stakeholders, campaign outcomes, and facts so no single type dominates
2. Quarter-aware -- when Business OS cadence is configured, current-quarter memories get higher weight
3. Recency-biased -- recent memories rank higher when similarity scores are close
4. Topic-focused -- when you provide topics, those drive additional semantic searches beyond the base retrieval
5. Reranked -- topic-driven candidates are over-fetched and reranked before the top context is passed into the brief
6. Source-exact for tasks -- when a meeting log is known, linked tasks and commitments are pinned ahead of semantic matches so action items are not lost

The brief is generated server-side (not by the LLM), then Claude polishes it for presentation.

Learned Tips

When you've been using Armbrain for a while, it learns patterns from your work -- what approaches worked, what went wrong, what to watch out for. These are extracted weekly as "tips" and automatically appended to your meeting briefs.

Tips appear at the end of the brief in two groups:

• Tips -- What works: Up to 3 strategy insights that have proven effective (e.g., "Acme responds well to competitive benchmarking in QBRs")
• Tips -- Watch out: Up to 3 warnings about past failures or pitfalls (e.g., "Avoid proposing paid social without ROI data -- rejected twice")

Tips come from your personal mind and are matched to the client you're prepping for. You don't need to configure anything -- they accumulate naturally as you use the system. If no tips exist yet, the section simply doesn't appear.

Output Reflection

Briefs are automatically validated against the client's most recent memories (last 14 days) before delivery. If Armbrain detects that something in the brief contradicts recent information or references stale data, it flags the issue in a reflection field.

This is fully automatic -- you don't need to enable or configure it. If everything checks out, no reflection is shown. If something looks off, you'll see it called out so you can address it before the meeting.

When to Use Meeting Prep

Key Details

• Prepping is read-only. Generating a brief never changes your data -- it only reads your memories and Brand DNA.
• Sensitive memories are restricted. Anything marked confidential or owner-only stays out of teammate views and client-facing outputs.
• The limit parameter controls how many memories are considered (default 50). Increase it for clients with extensive history.
• Brand DNA is always included. The brief shows stakeholders and brand voice regardless of topic.