knowledge

Mnemonic Knowledge Base

You have access to an organizational knowledge base with semantic search across 5 domains.

The Onion Principle

Always use the lightest layer first to minimize token consumption:

Layer	Tools	Tokens	When to use
0	search_quick, browse, count	~50/result	"What exists?" — scan first
1	search, search_*	~200/result	"What's relevant?" — semantic match
2	get_entity	~500-2000	"Tell me everything" — only specific items
3	find_related	N × ~50	"What's connected?" — graph traversal

NEVER request Layer 2 for multiple entities at once. Start with Layer 0/1, then drill down.

5 Domains

• commercial: opportunities, proposals, clients, competitors, client communications, followups
• operations: projects, tasks, deliveries, timeline, quality, logistics
• financial: budgets, APU, procurement, invoices, margins, expenses
• engineering: architectures, equipment, standards, protocols, configs, concepts
• knowledge: lessons, decisions, conversations, agent outputs, patterns

Proactive Search (do WITHOUT being asked)

• Before quoting → search_commercial + search_engineering
• Before estimating → search_financial("APU similar")
• Client mentioned → search_quick(domain=commercial, client=X)
• Technical decision → search_engineering("previous architecture")

Proactive Save (do WITHOUT being asked)

• Technical decision made → save_entity(domain=engineering, type=architecture)
• Lesson learned → save_entity(domain=knowledge, type=lesson)
• Client email/communication → save_entity(domain=commercial, type=client_comm)
• Equipment selected → save_entity(domain=engineering, type=equipment)
• Proposal generated → create_reference(ref_type=proposal) + save_entity