Skip to main content

Run the golden journey

After install, every organisation runs the same journey once before anything else: create the lanes beside production, give teams reach into them, build an agent in the lowest lane, test it, promote it lane by lane, and put it into production with a second person's approval. This page walks that journey in order and links the reference page for each step, so that one afternoon proves the environments, the roles and the release gate together.

Who can use it: the whole journey needs four people, or four accounts: an Admin or Super Admin for the lanes and the teams, a Builder in the development team, a tester in each higher lane's team, and an approver who holds the Admin tier or the Release Approver grant and is not the Builder.

The journey​

  1. Add the lanes

    As the Admin, open Environments and add the lanes beside production, for example a development lane, a test lane and a user-acceptance lane, each with its hostname, its size as a share of production and its profile. Watch each reach Active. Then set the order a version travels: development, test, acceptance, production. Add a lane walks the dialog and Operate lanes the statuses.

  2. Give teams reach

    Create one team per lane and grant each team its lane on the team's Access tab; a production lane needs a Super Admin. Put two people in each team, and add one approver at the Admin tier. Then prove the reach: each person signs in at their own lane and is admitted, and is refused at another lane. A Member cannot open the Control Room pages in production. Teams and The environment journey say what reach means and when it is enforced.

  3. Find the production model

    Decide how the lanes reach the model served in production. The supported route is to share the production deployment into the lane's gateway, where it appears as a read-only entry marked as shared from production; the lane's own provider key is the alternative. Operate lanes covers a lane's credentials.

  4. Build in the lowest lane

    As the Builder, in the development lane, add a knowledge source with two documents, a guardrail profile, a policy and a tool, then build an agent that binds all four and the production model. Publish it in the lane and chat with it there: the answer cites the documents, the tool runs, a prompt written to trip the guardrail trips it, and the policy applies.

  5. Promote into the test lane and test there

    As the Builder, request promotion into the test lane. A promotion into a non-production lane goes through at once. As a tester in that lane's team, chat with the promoted version and record what travelled with it, which is the agent's bindings, and what had to be re-created in the lane, which is the knowledge content and the credentials, because those belong to each environment. The environment journey explains the bindings.

  6. Promote into the acceptance lane

    Repeat the promotion into the user-acceptance lane and have that lane's tester run the same checks.

  7. Promote into production with an approver

    Request the promotion into production. It waits in Approvals for a reviewer other than the requester; the Builder cannot approve their own request. As the approver, read the platform's comparison of the version against what production serves and approve. Confirm the version serves, roll back once to the previous version and forward again. Review promotions and Publish, promote and roll back cover each dialog.

  8. Read the trail

    Open the Audit Log and filter on each person in turn. Every publish, promotion request, approval and rollback is there with the person who did it, and so is every lane created and every team granted. Audit Log lists what is recorded.

What comes after​

The journey above is the first day. These are the situations an organisation meets in the weeks after it, each proving one more thing about the same lanes and roles. Run them as you reach them.

ScenarioWhat it proves
A leaver mid-flow: the Builder who owns the agent and its knowledge source is offboarded with an heirThe heir keeps promoting; everything transfers; the leaver's key stops
Reach or a grant removed while someone worksThe next request is refused, not the next sign-in
A rejected promotion, and one that fails its gate; fix in the development lane and re-proposeThe ladder refuses with the reason named, and both decisions are in the audit log
Secrets and documents do not travel with a promotionA promoted agent without them fails with a named cause, never silently
A hotfix while lanes hold different versionsEach lane serves the version promoted to it, and rollback targets are the right ones
A new person arrives through single sign-on with no teamThe default team's reach applies, and a group rule lands them in the right team at the next sign-in
Resize, upgrade and delete a lane that holds a promoted versionNothing serving is lost, and each refusal names its reason
An auditor reads the whole trail across lanes, refused actions includedOne search finds everything one person did
A Super Admin overrides a stuck approvalThe override is recorded as such
Cost and monitoring per lane while every lane uses the production modelThe lane is a filter on every usage and cost view
A workflow agent with a tool-approval gate promoted the same wayThe ladder treats every kind of agent alike

Describes the platform as of .

Was this page helpful?