Model Lab — TypeScript SDK

Use your deployed API Router Gateway for model requests. Check the deployed version and required permissions. Example addresses are placeholders.

Inference playground: list callable models, create chat sessions, and stream responses.

Overview

The Model Lab resource is the inference playground. List callable models with a 16-field filter DTO, mint a session UUID for multi-turn conversations, and stream chat completions over Server-Sent Events as an async generator of raw payload strings.

Accessed via client.modelLab.

Available Operations

Method	Description
list()	List callable models
createSession()	Create a new chat session
stream()	Stream a chat completion over SSE

list

List callable models. pageSize / pageNum are required by the backend (code=1000 if missing) — the SDK defaults them to 20 / 1.

Example Usage

import { WltClient } from 'wlt-platform';

const client = new WltClient({ apiKey: 'your-api-key', baseUrl: 'https://console.example.com' });

// List inference models
const models = await client.modelLab.list({ usageType: 'inference' });
for (const m of models) {
  console.log(m.modelName, m.provider);
}

// Filter by provider and capability
const filtered = await client.modelLab.list({
  provider: 'dashscope',
  capability: 'text_to_text',
  inputs: ['text'],
  outputs: ['text'],
});

Parameters

Parameter	Type	Required	Default	Description
modelName	string	No	—	Model name (fuzzy search)
provider	string	No	—	Provider identifier
seriesProvider	string	No	—	Series provider identifier
modelType	string	No	—	Model type
viewAllFlag	number	No	—	View all flag (1=view all)
pageSize	number	No	20	Items per page. Required by backend; SDK defaults to 20
pageNum	number	No	1	Page number. Required by backend; SDK defaults to 1
usageType	string	No	—	Usage type: "inference" or "train"
trainingMethod	string	No	—	Training method: "sft" or "dpo"
tags	string[]	No	—	Tag filter list
capability	string	No	—	Capability filter, e.g. "text_to_text"
modelTypeList	string[]	No	—	Model type list (multi-select)
originProviders	string[]	No	—	Origin provider list (multi-select)
inputs	string[]	No	—	Input type list, e.g. ["text"]
outputs	string[]	No	—	Output type list (max 1 item), e.g. ["image"]

Response

Returns ModelInfoVO[]. See Model Gallery page() for field definitions.

Errors

Code	Error	When
1000	System error	pageNum / pageSize missing (backend requirement)
2000 / 2002	Authentication failure	API Key invalid
2007	Permission denied	Token lacks permission
3001	Token expired	Refresh failed

createSession

Create a new chat session. The returned UUID can be passed to stream() as sessionId for multi-turn conversations.

Example Usage

const sessionId = await client.modelLab.createSession();
console.log(`Session ID: ${sessionId}`);

Parameters

None.

Response

Returns string — a UUID session ID.

Errors

Code	Error	When
2000 / 2002	Authentication failure	API Key invalid
2007	Permission denied	Token lacks permission
3001	Token expired	Refresh failed

stream

Call a model via Server-Sent Events. Returns an async generator yielding the raw payload string of each data: line. The [DONE] sentinel terminates the iterator.

Example Usage

import { WltClient } from 'wlt-platform';

const client = new WltClient({ apiKey: 'your-api-key', baseUrl: 'https://console.example.com' });

// Create a session for multi-turn conversation
const sessionId = await client.modelLab.createSession();

// Stream inference. Use a provider + model actually served by the gateway --
// discover via `client.modelLab.list()` / `client.usage.gatewayProviders()`.
// On the daily environment, `openrouter` + `openai/gpt-5.4-nano` is a known-good pair.
for await (const chunk of client.modelLab.stream({
  provider: 'openrouter',
  modelName: 'openai/gpt-5.4-nano',
  userPrompt: 'Explain quantum computing in one paragraph.',
  sessionId,
  options: {
    temperature: 0.7,
    maxToken: 2048,
    topP: 0.9,
    systemPrompt: 'You are a helpful assistant.',
  },
})) {
  try {
    const evt = JSON.parse(chunk);
    const delta = evt?.choices?.[0]?.delta?.content;
    if (delta) process.stdout.write(delta);
  } catch {
    // Non-JSON chunk (e.g. heartbeat) — write through raw
    process.stdout.write(chunk);
  }
}
console.log(); // newline after streaming completes

Parameters

Parameter	Type	Required	Default	Description
provider	string	Yes	—	Provider identifier
modelName	string	Yes	—	Model name
userPrompt	string	Yes (unless retry)	—	User prompt text
sessionId	string	No	—	Session ID (UUID) for multi-turn conversations
taskId	string	No	—	Task ID — references the failed task when retrying
taskType	string	No	"TEXT"	Task type
isRetry	boolean	No	false	Whether this is a retry request (requires taskId)
options	OptionsDTO	No	—	Model parameter configuration (see below)

OptionsDTO:

Field	Type	Required	Default	Description
temperature	number	No	—	Temperature; controls output randomness
maxToken	number	No	—	Maximum generated token count
topP	number	No	—	Top-P sampling parameter
topK	number	No	—	Top-K sampling parameter
presencePenalty	number	No	—	Presence penalty
frequencyPenalty	number	No	—	Frequency penalty
stopSequences	string	No	—	Stop sequences
systemPrompt	string	No	—	System prompt
enableThinking	boolean	No	—	Enable deep thinking
enableProgress	boolean	No	—	Enable progress push
timeoutSeconds	number	No	—	Task timeout (seconds)
enableCache	boolean	No	—	Enable result caching
enableDocumentInlining	boolean	No	—	Enable document inlining
budgetTokens	number	No	—	Budget token count

Response

Returns AsyncGenerator<string, void, undefined>. Each yielded value is the string after data: on one SSE event; each chunk follows the OpenAI-compatible streaming format. The generator finishes when data: [DONE] is received or the connection closes.

Errors

Code	Error	When
1001	Validation error	provider / modelName missing, or isRetry=true without taskId
2000 / 2002	Authentication failure	API Key invalid
2007	Permission denied	Token lacks permission
3001	Token expired	Refresh failed