@osdk/aip-core

Core building blocks for AIP applications backed by the Foundry Language Model Service (LMS). Provides streaming chat completions, tool calling, and an AI SDK adapter.

Installation

pnpm add @osdk/aip-core

Peer dependencies:

pnpm add @osdk/client @osdk/language-models

Native API

foundryModel

Creates a language model handle for Foundry LMS.

import { foundryModel } from "@osdk/aip-core";

const model = foundryModel({
  client, // PlatformClient from @osdk/client
  model: "gpt-4o", // LMS model API name or a ModelIdentifier
});

The model parameter accepts either a string (LMS API name like "gpt-4o") or a structured identifier:

// Built-in LMS model
{ type: "lmsModel", apiName: "gpt-4o" }

// Customer-registered model
{ type: "registeredModel", registeredModelRid: "ri.language-model-service..." }

generateText

Performs a single chat completion and returns the full result.

import { foundryModel, generateText } from "@osdk/aip-core";

const model = foundryModel({ client, model: "gpt-4o" });

const result = await generateText({
  model,
  prompt: "Summarize the key points of this document.",
});

console.log(result.text);
console.log(result.usage); // { inputTokens, outputTokens, totalTokens, ... }

With system instructions

const result = await generateText({
  model,
  system: "You are a helpful assistant that responds concisely.",
  messages: [
    { role: "user", content: "What is TypeScript?" },
  ],
});

With tool calling

const result = await generateText({
  model,
  prompt: "What's the weather in London?",
  tools: {
    getWeather: {
      description: "Get the current weather for a location",
      inputSchema: {
        type: "object",
        properties: {
          location: { type: "string" },
        },
        required: ["location"],
      },
    },
  },
  toolChoice: "auto",
});

for (const call of result.toolCalls) {
  console.log(call.toolName, call.input);
}

Callbacks

const result = await generateText({
  model,
  prompt: "Hello",
  onStepFinish: (step) => {
    console.log("Step finished:", step.finishReason);
  },
  onFinish: (event) => {
    console.log("Total usage:", event.totalUsage);
  },
});

Sampling parameters

| Parameter | Description | | ------------------ | -------------------------------- | | maxOutputTokens | Maximum tokens to generate | | temperature | Randomness (0 = deterministic) | | topP | Nucleus sampling threshold | | presencePenalty | Penalize repeated topics | | frequencyPenalty | Penalize repeated tokens | | stopSequences | Stop generation at these strings | | seed | Deterministic sampling seed |

AI SDK Provider

@osdk/aip-core ships a custom provider for the AI SDK, allowing you to use Foundry LMS models with generateText, streamText, and other AI SDK functions.

Additional dependencies

pnpm add ai @ai-sdk/provider

Usage

import { createFoundryAI } from "@osdk/aip-core/ai-sdk";
import { generateText } from "ai";

const foundryAI = createFoundryAI({ client });

const { text } = await generateText({
  model: foundryAI("gpt-4o"),
  prompt: "Hello, world",
});

Streaming

import { createFoundryAI } from "@osdk/aip-core/ai-sdk";
import { streamText } from "ai";

const foundryAI = createFoundryAI({ client });

const result = streamText({
  model: foundryAI("gpt-4o"),
  prompt: "Write a poem about TypeScript.",
});

for await (const chunk of result.textStream) {
  process.stdout.write(chunk);
}

Registered models

const model = foundryAI({
  type: "registeredModel",
  registeredModelRid: "ri.language-model-service...",
});

Tool calling with AI SDK

import { createFoundryAI } from "@osdk/aip-core/ai-sdk";
import { generateText, tool } from "ai";
import { z } from "zod";

const foundryAI = createFoundryAI({ client });

const result = await generateText({
  model: foundryAI("gpt-4o"),
  prompt: "What's the weather in London?",
  tools: {
    getWeather: tool({
      description: "Get the current weather for a location",
      parameters: z.object({ location: z.string() }),
      execute: async ({ location }) => {
        return { temperature: 18, condition: "cloudy" };
      },
    }),
  },
});

v0 Limitations

The current release (v0) has the following limitations. Unsupported options surface as warnings in the result rather than throwing errors.

• Single-step only — multi-step tool loops (stopWhen, prepareStep) are not yet supported
• Text content only — image and file parts in messages are ignored with a warning
• No structured output — JSON response format is not yet wired
• No topK — the Foundry LMS OpenAI proxy does not accept this parameter

License

Apache-2.0