Concepts
Every other page assumes these. Each entry links to the page that covers it properly.
Routing
Section titled “Routing”Route. A name an engineer’s tool asks for in place of a model, such as
standard or deep. An administrator decides which model the name runs
on and who may use it. The list of routes is the whole of what a deployment
permits: a name that is not a route is refused, and that includes a real model
id. See Routes.
Model provider. Where models are invoked. There are two kinds,
bedrock (Amazon Bedrock’s runtime endpoint) and bedrock-mantle (Bedrock’s
OpenAI-compatible endpoint), and each is set to one AWS region. See
Model providers.
Catalogue. The models your AWS account can actually invoke, which Panorama discovers by asking each provider. A route can only point at a model in the catalogue.
Inference profile. The prefix on a Bedrock model id, such as us., au.
or global.. It decides which countries a request may be processed in and what
it costs, so us.anthropic.claude-sonnet-5 and
global.anthropic.claude-sonnet-5 are two different catalogue entries.
Spending
Section titled “Spending”Budget. A spending limit on a person, a team or the whole organization, over a day, a week or a month. See Budgets.
Soft limit. The warning line. Past it, requests are still served, except on routes you marked as stopping there. The engineer is told which routes still work and switches.
Hard limit. The stop line. Past it, every request governed by that budget is refused until the window resets or somebody raises the limit.
Unpriced. A request whose model has no published rate. It is recorded with no cost and counted separately, never as $0.00.
Attribution
Section titled “Attribution”Work item. The thing a request paid for: a branch, a pull request or a ticket. See How attribution works.
Session. One conversation in a coding tool. Attribution is decided once per session and every request in it inherits the answer.
Provenance. How Panorama knows a request belongs to a work item: the client said so (stamp), it was read from the request’s structure (structural), or a ticket key was mentioned in the text (content). It is shown beside every attribution.
Unattributed. Spend that could not be tied to a work item. It is always shown next to the attributed figure.
Cost center. An accounting line, such as CC-100 Payments. Rules map a
repository or a ticket prefix to one. It reports and warns; it never blocks.
See Cost centers.
People
Section titled “People”Role. admin, manager or developer. An admin configures everything. A
manager sees the totals for teams they lead. A developer sees their own usage
and nothing else. Roles come from your identity provider’s groups.
Team. A group of people whose spend rolls up together, and the smallest unit a manager is shown.
Credentials
Section titled “Credentials”You can tell them apart by prefix. See Credentials.
| Prefix | What it is | Who holds it |
|---|---|---|
pnt_ |
Gateway token | An engineer’s coding tool |
pnk_ |
Service token | A CI pipeline, billed to a team |
pnb_ |
Setup code | Whoever is setting the deployment up, once |
Setup session. What a setup code opens: an hour in which you can configure single sign-on and the model providers, and nothing else. It is not an account.
This manual is licensed underCC BY 4.0. Code samples in it are part of Panorama, licensed under theGNU AGPL v3.0 only.