TramAI - governed AI workflows for Java and Kotlin

tramai-spring-boot-starter-local-provider-openai

Version: 0.6.0
Status: Preview
Role: Local (non-egress) provider starter for development, smoke tests and offline labs.

Purpose

This starter registers an OpenAI-compatible ModelProvider pointing at a local endpoint — Ollama, llama.cpp, vLLM or any OpenAI-compatible server on the machine — from a single YAML entry. It exists so a sovereign or offline deployment can run real inference without adding a cloud provider adapter.

It is deliberately narrow: only the local-lab-provider entry with type: openai triggers bean creation. Any other key, or a different type, is skipped silently.

What it provides

ComponentDetail
OpenAiCompatibleProviderAutoConfigurationRegistered through META-INF/spring/...AutoConfiguration.imports
TramaiProviderPropertiesBinds tramai.providers.* entries (type, base-url, api-key, model)
localLabModelProvider beanAn OpenAiCompatibleProvider with provider name local-lab-provider

Behavior:

  • tramai.providers.local-lab-provider.base-url is required; without it startup fails.
  • api-key falls back to local-dev, which is enough for servers that do not check credentials.
  • The bean is @ConditionalOnMissingBean(name = "localLabModelProvider"), so your own bean wins.

Dependencies

dependencies {
    implementation(platform("dev.tramai:tramai-bom:0.6.0"))
    implementation("dev.tramai:tramai-spring-boot-starter")
    implementation("dev.tramai:tramai-spring-boot-starter-local-provider-openai")
}

tramai:
  profile: sovereign
  providers:
    local-lab-provider:
      type: openai
      base-url: http://localhost:11434/v1
      api-key: local-dev
      model: qwen2.5:7b
  sovereign:
    allowed-models: [local-lab-model]
    allowed-providers: [local-lab-provider]
    provider-zones:
      local-lab-provider: LOCAL
    models:
      local-lab-model: local-lab-provider

When to use this starter

  • You want a real model on the loopback interface while developing or running an offline lab.
  • You are exercising the sovereign runtime against a local OpenAI-compatible server.
  • You need a provider that cannot accidentally reach the public internet.

When NOT to use this starter

  • Production traffic — pin the provider you actually run against, through tramai-spring-provider-openai or tramai-spring-provider-ollama.
  • You need more than one property-configured provider: only local-lab-provider is wired by this starter (v1 scope), so additional providers need their own adapter module or bean.
  • You expect it to enforce no-egress guarantees: it points at a local URL, but network isolation is an infrastructure concern.
Previous
Next