WireAIProvider

The root context provider. Place it once above any component that uses wireai-rn hooks. It initializes the component registry, LLM adapter, and shared context.

Usage

App.tsx

1import { WireAIProvider, defaultComponents } from "wireai-rn";
2
3<WireAIProvider
4  llm={{ provider: "ollama", baseUrl: "http://localhost:11434", model: "llama3" }}
5  components={defaultComponents}
6  systemPromptSuffix={appInstructions}
7  initialMessages={restoredHistory}
8  maxContextMessages={20}
9  maxContextChars={12000}
10  onThreadUpdate={(msgs) => saveToStorage(msgs)}
11>
12  {children}
13</WireAIProvider>

Props

Prop	Type	Default	Description
`llm`required	`LocalLLMConfig`	—	LLM provider configuration. See LLM Adapters.
`components`required	`WireAIComponent[]`	—	Components the LLM can render. Use `defaultComponents` or mix with your custom ones.
`systemPromptSuffix`	`string`	undefined	Appended verbatim to the auto-generated system prompt. Use for app persona, conversation flows, or domain-specific rules.
`initialMessages`	`Message[]`	[]	Pre-populate the conversation (e.g., restored from storage on app launch).
`maxContextMessages`	`number`	20	Maximum number of messages sent to the LLM per request. Oldest messages are trimmed first.
`maxContextChars`	`number`	12000	Maximum total characters sent to the LLM. Oldest messages are trimmed first. ~12 000 chars ≈ 3 000 tokens.
`onMessage`	`(msg: Message) => void`	undefined	Called each time a user or assistant message is added to the thread.
`onThreadUpdate`	`(msgs: Message[]) => void`	undefined	Called with the full history whenever it changes. Ideal for persistence.
`licenseKey`	`string`	undefined	Reserved for future use. Has no effect in v0.x.

LocalLLMConfig

types

1type LocalLLMConfig = {
2  provider: "ollama" | "lmstudio" | "openai" | "webhook" | "custom" | "a2a";
3  baseUrl: string;    // e.g. "http://localhost:11434" or "https://api.openai.com"
4  model: string;      // e.g. "llama3" or "gpt-4o-mini"
5  apiKey?: string;    // required for OpenAI; optional for Webhook and A2A
6  temperature?: number;
7  maxTokens?: number;
8  timeoutMs?: number; // default: 60 000ms
9};

Provider remounting

When the user switches LLM providers (e.g., in a settings panel), you want the adapter, system prompt, and message history to reset completely. Use a key prop derived from the config:

Tsx

1<WireAIProvider
2  key={`${config.provider}:${config.baseUrl}:${config.model}`}
3  llm={config}
4  components={components}
5>
6  {children}
7</WireAIProvider>

This pattern is used in the mental-coach example to reset the full conversation when switching between Ollama, OpenAI, and A2A providers.

Context budget

wireai-rn trims older messages when the conversation grows beyond maxContextMessages or maxContextChars. The system prompt is always sent regardless of budget. Tune based on your model's context window:

Model	Recommended maxContextMessages	Recommended maxContextChars
GPT-4o / Claude 3	30+	40 000+
GPT-4o-mini	20	12 000
Llama 3 8B (Ollama)	10	6 000
Phi-3 mini (Ollama)	6	3 000

Dev-mode guardrails

In __DEV__ mode, the provider automatically logs warnings for common misconfigurations:

⚠API keys in config: API keys should never be bundled in a mobile app. Use the Webhook adapter for production.
⚠Large registries (> 10 components): Local models (Llama 3, Phi-3) work best with fewer components. JSON quality degrades with large registries.
⚠LLM connectivity: Performs a HEAD request to baseUrl on mount to verify the server is reachable.

← Quick Start Hooks →