Model APIs¶

With the agent loop on the device, ai-elements-core calls model APIs directly. Each is a ChatBackend with tools, reasoning, images and usage:

ProviderProfile describes a provider (kind, base URL, model) and builds its backend or its model binding for the harness:

val profile = ProviderProfile("openai", "OpenAI", ProviderKind.OPENAI_RESPONSES, "https://api.openai.com/v1", "gpt-5.5")
val backend = profile.createBackend(apiKey = key, approver = approver)   // built-in agent loop
val model = profile.model(apiKey = key)                                  // for AgentHarness

profile.listModels(apiKey) asks the provider which models it serves, for a model picker.

Keys and sign-in¶

ProviderStore keeps API keys encrypted with an Android Keystore key (SecretStore); they are never logged.

Its built-in profiles are ProviderProfile.Presets (OpenAI, Anthropic, Gemini, OpenRouter and the offline demo agent). Add your own agent servers to them: ProviderStore(context, ProviderProfile.Presets + myServers). The demo adds the bundled reference server over each protocol this way (DemoPresets).

Some providers support signing in instead of an API key, with OAuth 2.1 (PKCE S256, a loopback redirect on 127.0.0.1 or the device flow, RFC 8628): OpenRouter, and, experimentally, ChatGPT (Codex), xAI and Kimi Code subscriptions. Tokens are refreshed automatically.

Koog¶

ai-elements-koog runs a JetBrains Koog agent as a ChatBackend, or as the harness's model binding, with AI Elements tools as Koog tools. Approvals and progress work as with the built-in loop.