Memory Layer Standardization Audit (Onboarding)

Memory Layer Standardization Audit (Onboarding)

1. 01

   Inventory every AI tool and agent your agency runs for the client

   List which tools hold client context: coding agents, chat assistants, and automation workflows. Note which ones already persist memory and which start fresh each session.

2. 02

   Map the client knowledge that must persist across sessions

   Identify non-negotiable context: brand guidelines, technical architecture, past decisions, and stakeholder preferences. This is the minimum set your memory layer must retain.

3. 03

   Evaluate each memory tool's export and portability options

   Check whether the tool can export notes or graph data in an open format like JSON or Markdown. If export is locked, that vendor lock-in becomes a real risk to your institutional knowledge.

4. 04

   Compare graph-based memory against structured project archives

   Graph approaches like Cogni's entity-graph spreading activation excel at multi-hop reasoning, while structured archives like Knownbase organize notes by project and tag. Choose based on whether your workflows need associative recall or simple retrieval.

5. 05

   Test retrieval accuracy with a sample of client-specific queries

   Run five to ten realistic queries that require cross-session context, such as 'What was the agreed CTA for the Q3 campaign?' Measure how often the memory layer returns the correct answer without re-prompting.

6. 06

   Check for shared access across your agency's agent stack

   Confirm the memory layer connects to the specific tools your team uses, whether that is Claude Code, ChatGPT, Cursor, or others. A memory silo that only one tool can read defeats the purpose.

7. 07

   Document the chosen memory layer and its governance rules

   Write down who can write to the memory, what gets stored, and how stale entries are reviewed. This prevents the memory from becoming a dumping ground for outdated client facts.