Useful context.
Clear boundaries.
Trust starts with knowing what is being searched, who can retrieve it, and where it runs. For companies, teams, and people, these are decisions to agree and test in your evaluation.
Start with code and saved knowledge. Source access, hosting, and data handling are confirmed for your deployment.
Agree it.
Then test it.
An evaluation should include the questions that must return useful context and the cases that must stay outside its audience.
- Sources
Choose what enters the evaluation.
Name the repositories and saved knowledge you want to search. Confirm integration support, source permissions, and how changes or removals are handled.
Read the source guide- Audience
Make personal and shared scope explicit.
Agree who can retrieve sources, recall findings, and see shared activity. Test the intended audience, a person outside that audience, and what happens when access changes.
- Environment
Know where the work happens.
Confirm where ShepAIrd runs, how people and agents sign in, and which model providers your deployment uses. Confirm host compatibility for the evaluation.
- Data handling
Set terms before connecting data.
Agree storage, retention, deletion, and provider handling for the evaluation. The website’s privacy policy describes the site and evaluation correspondence; your deployment needs its own agreed terms.
Read the website privacy policy
Evidence should
be open to inspection.
Context search brings supporting evidence into view. Memory retains a reusable finding with its origin and freshness so you can inspect the evidence before acting.
Include stale, changed, and revoked sources in your trial. Check what people and agents can recall after those changes, alongside the questions you already know how to answer.
Inspect the example sources