Model services, accounts, and routes
Configure a model request path and understand every name in the selector.
Morphz models access through wire protocols, service adapters, provider instances, auth accounts, and model routes. Keeping them separate prevents the runtime from becoming coupled to one vendor.
Five separate objects
| Object | Question it answers |
|---|---|
| Protocol | What are the HTTP and stream semantics? |
| Provider Adapter | Which endpoints, headers, and dialect does the service require? |
| Provider Instance | Where does this deployment connect? |
| Auth Account | Which login identity or API key is used? |
| Model Route | How does a user selection resolve to a physical model and account? |
Supported protocols
The current wire protocols are OpenAI Responses, OpenAI Chat Completions, Anthropic Messages, and Gemini generateContent. Compatible gateways, self-hosted services, and API-key endpoints reuse these protocols instead of introducing a domain model for every brand.
Physical names and aliases
A physical model name comes from the service catalog or explicit operator input. Morphz does not guess model names from a brand and does not create plausible-looking defaults.
A route may define display_alias. The selector displays that alias when present; otherwise it displays an explainable physical name. A generated route ID remains an internal stable reference unless the operator explicitly records it as an alias.
Direct route target
[models.my-coding-model]
display_alias = "Coding"
service = "codex-subscription"
physical_model = "gpt-example-coding"
account = "account-example"
The physical name above is a structural example and must be replaced with a model the service currently accepts.
Multiple candidates
A route can contain multiple ordered candidates when several services provide the same user-facing choice. The runtime selects one eligible, enabled, healthy candidate. Account cooldown or network failure may move selection to another candidate; one request does not use multiple accounts simultaneously.
Model capacity
- Context window: combined input and generated output limit;
- Maximum input: prompt-only limit;
- Maximum output: generation allowance.
Morphz stores capacity fields returned by the service and leaves absent fields unknown. Operators may override a confirmed limit; the Runtime never fills missing fields with guessed values.