vertex-doctor walkthrough

Probe, diagnose and unblock model access on Google Cloud Agent Platform (formerly Vertex AI). Recorded against real projects (one created minutes before recording, one sandbox); identifiers are redacted. Start with the journey.

Short clips

vertex-doctor journey for a new Claude lineage: nine steps green, step 8 marked [you] with the quota request form ready to paste

journey: one model, one location, the same ten steps every time. Here a brand-new Claude lineage stops at the one step a person has to do (quota the automated adjuster refuses), with the form to send.

vertex-doctor journey for gemini-flash-latest: all ten steps green, then the curl and SDK call that work

journey on the happy path: all ten steps green, then the curl that worked and the official SDK call; other clients on request.

vertex-doctor quota-plan: the quota bundle sized from --rpm, the base_model bucket, denied requests already on file, and a filled request form

quota-plan: the bundle (QPM, input and output tokens/min) sized from --rpm, the base_model bucket QuotaInfo actually keys on, requests already on file and their outcome, and a filled request form.

vertex-doctor probe: a matrix of models by location with one fix per cause

probe: many models x locations, one tiny request per cell, classified into a fixed taxonomy, one command per cause.

vertex-doctor doctor --fix: environment, probe, fixes, re-probe, before and after

doctor --fix: the matrix version of the loop: environment, probe, fix plan, re-probe of the corrected cells, before -> after.