@onioneko/kiki-llm-openai

Kiki · Middleware framework for LLM agents

npm install @onioneko/kiki-llm-openai

OpenAI-compatible LLM provider as middleware. Calls the Chat Completions API and pushes the assistant response onto ctx.messages, then calls next() to allow response-processing middleware downstream.

Compatible with any OpenAI-compatible provider (OpenAI, Azure OpenAI, Ollama, Together, vLLM, etc.) via the baseURL config.

Usage

import { compose } from "@onioneko/kiki-core";
import { createOpenAIMiddleware } from "@onioneko/kiki-llm-openai";

const run = compose([
  // ... other middleware ...
  createOpenAIMiddleware({ model: "gpt-4.1" }),
]);

await run({ messages: [{ role: "user", content: "hello" }] });

With a compatible provider

createOpenAIMiddleware({
  baseURL: "http://localhost:11434/v1", // Ollama
  model: "llama3",
});

Configuration

model is required. Other fields are optional with sensible defaults.

| Option | Env var | Default | Description | |------------|------------------|-----------|------------------------------------------| | model | — | — | Model identifier (required) | | apiKey | OPENAI_API_KEY | — | API key (required unless client injected) | | maxTokens| — | 16384 | Maximum tokens in response | | baseURL | OPENAI_BASE_URL| — | Base URL for compatible providers | | client | — | — | Injected OpenAI client (for testing) |

Config precedence: explicit config > default.

Behavior

• Non-terminal — calls next() after pushing the response, allowing response-processing middleware to be composed after the LLM.
• Multi-modal — handles text and image content blocks (images sent as data URIs).
• System prompt — ctx.system is prepended as a role: "system" message.
• OpenAI-compatible — works with any provider that supports the Chat Completions API.

Exports

• createOpenAIMiddleware(config) — factory function returning a Middleware.
• OpenAIConfig — configuration type.
• Re-exports from @onioneko/kiki-types: Message, ConversationContext, ContentBlock, TextBlock, ImageBlock.