Architecture Questions
I'm trying to get the most out of QMD for long-term business / company knowledge in my agent's workspace and would love some advice from people who've scaled this successfully.Right now my main question is about the best architecture for important / frequently referenced context (company policies, key clients, product specs, pricing logic, internal playbooks, etc.):
Option A – Heavy use of pointers / routing index
Keep a clean, concise MEMORY.md (or INDEX.md / BUSINESS-CONTEXT.md) that acts purely as a table of contents / routing layer (~30–80 lines max).
Example:
→ Company Overview & Mission → overview.md
→ Product & Pricing → products-pricing.md
→ Key Client Profiles → clients/
→ Sales Playbooks → playbooks/sales-2026.md
→ Internal Tools & API Keys → tools.md (do not commit keys!)
→ Compliance & Legal Notes → compliance.md
Then rely on QMD's hybrid search (BM25 + vectors + re-ranking) to pull the actual detail files when needed.Option B – Dump more directly into MEMORY.md
Put the really critical / high-priority facts straight into MEMORY.md so they're almost always in context, and use separate .md files only for very long / detailed stuff.Option C – Hybrid / something else
Or do you do a mix? Multiple index files? Folders with one index per department? Symlinks? Naming conventions that help QMD rank better?Extra context:Using QMD backend (hybrid mode, not pure search)
Trying to minimize token waste and "forgetting" of important stuff
Want recall to be fast & precise even after months of sessions
Workspace isn't tiny anymore (~hundreds of md files)
What has worked best for you when the agent needs to reliably remember & apply important business rules/context without you re-explaining every few weeks?
Option A – Heavy use of pointers / routing index
Keep a clean, concise MEMORY.md (or INDEX.md / BUSINESS-CONTEXT.md) that acts purely as a table of contents / routing layer (~30–80 lines max).
Example:
Core Business Context – Last updated 2026-03
→ Company Overview & Mission → overview.md
→ Product & Pricing → products-pricing.md
→ Key Client Profiles → clients/
→ Sales Playbooks → playbooks/sales-2026.md
→ Internal Tools & API Keys → tools.md (do not commit keys!)
→ Compliance & Legal Notes → compliance.md
Then rely on QMD's hybrid search (BM25 + vectors + re-ranking) to pull the actual detail files when needed.Option B – Dump more directly into MEMORY.md
Put the really critical / high-priority facts straight into MEMORY.md so they're almost always in context, and use separate .md files only for very long / detailed stuff.Option C – Hybrid / something else
Or do you do a mix? Multiple index files? Folders with one index per department? Symlinks? Naming conventions that help QMD rank better?Extra context:Using QMD backend (hybrid mode, not pure search)
Trying to minimize token waste and "forgetting" of important stuff
Want recall to be fast & precise even after months of sessions
Workspace isn't tiny anymore (~hundreds of md files)
What has worked best for you when the agent needs to reliably remember & apply important business rules/context without you re-explaining every few weeks?
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
131,281Members
Resources
Similar Threads
Was this page helpful?
