ModelKit

ModelKit is the unified model gateway inside AgentOS.
If your app already has its own complex agent, or if you only need direct model access, this is usually the best place to start.

What it solves

Its main value is that developers do not have to solve model sourcing first.

It exposes model capabilities by task:

Task	Route prefix	Description
`chat`	`/modelkit/chat`	Text and vision models through the OpenAI chat completions protocol
`embedding`	`/modelkit/embedding`	Text embeddings
`asr`	`/modelkit/asr`	Speech-to-text
`tts`	`/modelkit/tts`	Text-to-speech

Best fit

you already own the agent runtime and only need the model-calling layer
you want to call models directly through the AgentOS SDK or OpenAI-compatible protocol
you want one gateway for text, multimodal, TTS, ASR, and embeddings

Most common usage

Use the AgentOS SDK

Both TypeScript and Dart SDKs expose ModelKit through sdk.modelkit.

import { AgentOSSDK, ModelTask } from 'agentos-sdk-ts';

const sdk = new AgentOSSDK({ baseUrl: 'http://127.0.0.1:8888' });
const chatModels = await sdk.modelkit.listModelsByTask(ModelTask.chat);

const completion = await sdk.modelkit.chat.create({
  model: chatModels[0].id,
  messages: [{ role: 'user', content: 'Hello from ModelKit.' }]
});

Use the OpenAI SDK directly

AgentOS task routes expose OpenAI-compatible protocols, so in many cases you can use the official OpenAI SDK directly. Make sure baseURL points to the specific task prefix.

Parameter mapping:

OpenAI SDK field	AgentOS source
`baseURL`	For chat, `${baseUrl}/modelkit/chat/v1`, where `baseUrl` comes from root `GET /`
`apiKey`	bearer token returned by `AppKit` registration
`model`	`model.id` from `sdk.modelkit.listModelsByTask(ModelTask.chat)`

Capability scope

chat completions: /modelkit/chat/v1/chat/completions
image-related model input: still use /modelkit/chat; check capabilities.vision
tts: /modelkit/tts/v1/audio/speech
asr: /modelkit/asr/v1/audio/transcriptions
embeddings: /modelkit/embedding/v1/embeddings

When ModelKit should not be your first stop

you need more than model calls and want the platform to orchestrate the agent and tool chain
you need ready-made RAG services rather than standalone embeddings

In those cases, start with:

ModelKit ​

What it solves ​

Best fit ​

Most common usage ​

Use the AgentOS SDK ​

Use the OpenAI SDK directly ​

Capability scope ​

When ModelKit should not be your first stop ​

Next steps ​