Client Context Ingestion Checklist (Onboarding)

Client Context Ingestion Checklist (Onboarding)

1. 01

   Inventory every client-specific fact the agent stack will need before the first prompt runs

   Brand guidelines, approved terminology, product architecture, compliance constraints, and named stakeholders. Missing facts get re-typed into prompts for months, so capture them once at kickoff.

2. 02

   Assign each fact a stable entity name and a project tag

   Entity-graph memory such as Cogni retrieves connected facts through spreading activation, which only works when the same client, product, or campaign is named consistently across notes.

3. 03

   Write the initial memory set through one connector, not five

   Pick a single write path for onboarding week. Knownbase and OzBrain both accept agent-written notes over MCP, but parallel writers during setup produce duplicate articles that later retrieval has to reconcile.

4. 04

   Record provenance on every entry: source document, date, and the person who approved it

   Jylus compiles Context Packs with proof IDs and temporal evidence. Mirror that discipline manually if your connector does not attach source metadata automatically.

5. 05

   Test recall with three questions a new account manager would ask

   Ask for the client's tone rules, their last major campaign decision, and one constraint that changed mid-project. If the memory layer cannot answer all three, the ingestion is incomplete.

6. 06

   Confirm the export format before you commit client knowledge to the layer

   Run a full export and open the file. If the output is a proprietary blob rather than readable text or JSON, the agency carries lock-in risk on institutional knowledge built over the retainer.

7. 07

   Set a review cadence for stale entries at 30 and 90 days

   Client brand rules and product lines change. Schedule a quarterly pass where a named owner marks entries as current, superseded, or archived.