Procedures
Troubleshooting is indexed by symptom: something is wrong, and you want to know what it means. These pages are the other half — indexed by task, for the jobs that have a right order and a check at the end.
Every procedure ends the same way: a step that proves it worked, and what to send us if it did not. None of them needs us to be reachable.
| Procedure | When |
|---|---|
| Prove the egress before you install | Once the firewall requests are said to have landed |
| Which Agent Router credential buys what | Before you build the Secret |
| Add a context stream from the shelf | Choosing a prebuilt context stream to add, or adding one of your own |
| Take a source live | Flipping GitHub, Jira or Router spend from synthetic to real data |
| Flip a source to the ingest engine | Moving a live GitHub or Jira source from its legacy lane onto the manifest ingest engine, one source at a time — and back |
| Pre-flight an Oracle view before you connect it | You are about to register a view over an Oracle database and want the first pass to work first time |
| Connect a Snowflake table | You have a table or view in Snowflake — accounts, incidents, changes, any gold-layer object — and want to ask questions about it, read in place |
| Connect your Snowflake directory | You have a Snowflake view of your own people and want per-person answers. The one source you author rather than configure |
| Read an Oracle view as a context stream | You have a view in an Oracle database — a resource plan, a portfolio, a timesheet extract — and want to write the document that reads it, in place or as dated copies |
| Set up a stream with the assistant (experimental) | You registered a stream and want an assistant to customise its recipe, or describe a query-through stream from scratch. Every proposal waits for an administrator's approval |
| Confirm story points are being read | After Jira goes live |
| Turn on email sign-in | Giving more than one person access, with no IdP in front |
| Sign in with Entra ID | Putting your own identity provider in front of Prism |
| Rotate a credential | Expiry, revocation, or a routine rotation |
| Upload a classification of your people | You keep the combinations that decide who counts as a developer in a spreadsheet, and want questions scoped to exactly that list |
| Widen a source's scope | You granted a credential more repositories or projects and want Prism to hold all of it |
| Your own name and logo | You want the appliance to carry your product name and mark rather than Prism's |
| Re-own the shared agent | Every chat turn fails with "it is owned by another identity" |
| Back up and restore | Before you need it. Your definitions are the one thing here nobody else has a copy of |
Two of these are worth doing before the thing they are about, rather than after it goes wrong:
- Prove the egress before you install. A missing firewall rule discovered on the first chat turn, or on the first ingest days later, presents as Prism being broken rather than as a route being closed.
- Take a source live is two upgrades, not one, because the flip is destructive: the first live ingest truncates the source's synthetic rows, so the credential is proved while there is still something to lose.