Status: in preparation
Prism is in preparation. The specification is written and phase 1, read-only follow-up, is under way. Nothing is available to use today, and what is listed below is what is planned, phase by phase.
What the customer sees
- What runs: the applications of the organisation, their state, their versions and their sleep state.
- Consumption and cost: provisional at first, measured by Prism itself until the metering of the platform exists. These amounts are estimates, never invoices: the invoice stays the one of the billing tool.
- Optimisation hints: figured rules, for example a memory size far above what the application uses, an application that could sleep, replicas that stay idle, slow starts or a high error rate, each with its estimated saving.
- Traffic: how often each application is called.
What the customer manages
- Git repositories and builds, with the Bup builder.
- Fermion pipelines, starting from the standard workflow of the platform.
- Domains and DNS, APIs and the products consumed.
- Collaborators and their roles, and the access of AI assistants through Gauge.
Pipelines from the standard workflow
The platform provides a standard Fermion workflow for build and deployment: clone, conformity checks, native build, wake test, deployment to the test environment, human approval, deployment to production, and rollback if production fails. Each step can be switched on or off by a variable, without changing the model. This part comes in phase 3.
Isolation and rights
- One organisation per customer: a person of one organisation never sees a resource of another, nor its existence; the answer is the same as for a resource that does not exist.
- Four roles of the customer: owner, developer, viewer and billing. Amounts are seen by the owner and by billing only.
- Least privilege: the token of the person opens Prism and nothing else. Prism calls the services of the platform with its own service token and filters everything by organisation; it has no free proxy to the platform.
- Every write is audited, and the owner can read the journal.
Phases
Phase 1: read-only follow-up (applications, consumption, hints, traffic), with sleep and rollback. Phase 2: repositories and builds. Phase 3: Fermion pipelines. Phase 4: domains, DNS, APIs and products. Phase 5: access, AI assistants and the real metering.
Subscription and customer consoles
A customer's subscription is a named plan of platform services. The plan decides which services the organisation can use and which consoles it can open; Prism shows the plan and gives the way to each console.
- A subscription is a named, versioned plan. The first one is Developer: gDown, Crystallon, Fermion, Top and Charm, each with its console.
- Developer bills nothing: it is the plan for local development and for trying Gluonify, offered on sandbox environments. Usage is still measured.
- Prism has a Subscription view: the plan with its name and version, each service with its state and the applications that use it, the quotas, and an "Open the console" button per service. The button is shown to owners, developers and viewers, and only for a console the plan opens.
- Paid plans will come later, as other plans with a price list. No price is published yet.
The consoles
Each console is the same interface as the operator's, served for customers on their own address, app.<zone>, under the path below, with its own sign-in client. A customer reaches only its own organisation's data.
- Prism (app.<zone>/): The customer console: applications, consumption, access and the subscription view.
- Charm (app.<zone>/charm/console/): Charm, realm console: the clients, users and roles of the organisation's own realm (org-<org>), with its own signing key and issuer, for the users of its applications. Owners and developers write, viewers read.
- Crystallon (app.<zone>/crystallon/): Crystallon: templates, images and fonts, generated code and rendering. Owners and developers write; viewers read and render. Never document conversion nor template installation.
- Fermion (app.<zone>/fermion/): Fermion: the executions of the organisation, read-only (list and follow). Customers cannot start, resume or delete an execution, nor change a workflow.
- Top (app.<zone>/top/): Top: the secrets of the organisation's applications, written without ever being read back, and the list of their names. Viewers see the names only. Never a value.
- gDown (app.<zone>/gdown/): gDown: the query and model pages on the organisation's own databases. Owners and developers read and write, viewers read. No schema change, no other database.
How access works
- Sign-in goes through Charm. A console token carries the organisation and customer roles only (owner, developer, viewer, billing); an operator account never gets one.
- The plan decides. Charm issues a console token only if the organisation's plan opens that console, and checks it again at every refresh: closing a console in the plan stops it within a minute. If the check cannot be made, no token is issued.
- The billing role has no console: the contract, usage and invoices are in Prism.
- Everything is bounded to the organisation: another organisation's object is never visible, and the consoles have no operator rights.
- The applications use the same services through their declared uses: a dedicated gDown database, Top secrets, the Crystallon and Fermion APIs, and Charm, where the application gets the issuer of its organisation's realm and never the platform's.
Status: the consoles and the realm are built and covered by unit tests; end-to-end verification on a full cluster is still to come.
Limits
Prism neither administers the nodes, the cluster or the stack, nor creates organisations: these remain operator tasks in Synchrotron. It does not give access to what is inside the applications of the customer. Amounts are provisional until the metering of the platform is delivered.