This API is used for communication with a large language model (LLM) and text translation (HTML payload is also supported) against an externally hosted service.
Transfers the specified question to the LLM in order to receive a response. You may send a single message (role: user) or multiples messages (role: system/assistant/user) using the same src as a single request to the LLM.
The service reassembles the messages with an equal src per session and sends the messages array: [message1, message2, ...] to the LLM. Very large messages which don't fit into a single Websocket message may be chunked
using the parameter complete to indicate wether or not the last message chunk was sent.
e.g.
{ "Question", "message": { role: "user", content: "What is innovaphone?" }, "seq": 0, "complete": true; "last": true, "src": "src0" }
or
{ "Question", "message": { "role": "system", "content": "You are very helpful" }, "seq": 0, "complete": true, "src": "src1" }
{ "Question", "message": { "role": "user", "content": "What is innovaphone?" }, "seq": 1, "complete": true, "src": "src1" }
{ "Question", "message": { "role": "assistant", "content": "Innovaphone is a telecomunications company!" }, "seq": 2, "complete": true, "src": "src1" }
{ "Question", "message": { "role": "user", "content": "How many people work there?" }, "seq": 3, "complete": true, "last": true, "src": "src1" }
{
mt: "Question",
message: { role: "user", content: "The actual question comes here..." },
seq: 0,
complete: true;
last: true;
}| string mt | message type [mandatory] |
| object message | the message to be transfered [mandatory] |
| number seq | the sequence number of the message starting from 0 [mandatory] |
| boolean complete | if the message is not chunked or the message is the last chunk of a chunked messages set to true [optional/mandatory] |
| boolean last | needs to be set to true for the last message to be sent [optional/mandatory] |
| number temperature | a number between 0 and 1 which determines the creativity of the response (defaults to 0) [optional] |
| number max_tokens | limits the length of the response by specifying the maximum number of tokens (words and punctuation) to generate [optional] |
| number top_p | controls the probability threshold for generating tokens, ensuring that the generated tokens have a cumulative probability of at least the specified value [optional] |
| number n | specifies the number of chat completion choices to generate for each input message [optional] |
| string or array of strings stop | a string or array of strings that, if present, stops the generation process when the model outputs this token [optional] |
| number presence_penalty | penalizes the model for generating tokens that are already present in the conversation history, encouraging new content [optional] |
| number frequency_penalty | penalizes the model for generating repetitive tokens, thus reducing repetition in the responses [optional] |
| array of objects tools | provides function specifications to enable models to generate function arguments and invoke functions as part of the response [optional] |
| string tool_choice | forces the model to use a specific function or not use any function at all [optional] |
| number seed | sets the random seed for reproducibility, ensuring consistent outputs for the same input parameters [optional] |
{
mt: "QuestionResult",
message: { role: "assistant", content: "The answer to your question is....!" }
}| string mt | message type (is "QuestionResult") |
| object message | the response object of the LLM |
{
mt: "QuestionResult",
message: { content: "The actual error message" },
error: true
}| string mt | message type (is "QuestionResult") |
| string object | English error message which can be shown to the User |
| bool error | labelling whether error was send. Only sent if true [optional] |
Send to obtain all the available LLMs of the model service provider
{
mt: "GetModels"
}{
mt: "GetModelsResult",
message: [{"id":"qwen2","object":"model","created":1756396382,"owned_by":"library"}, ... ]
}| string mt | message type (is "GetModelsResult") |
| array object | An array of objects of all the available models |
You can consume the API com.innovaphone.assistant and send request to it. Inside the callback you can access the response text via recv.msg.message.content
var assistantApi = start.consumeApi("com.innovaphone.assistant");
let messages = [];
messages.push({ role: "system", content: "You are very helpful!", complete: true, src:"src0" });
messages.push({ role: "user", content: "What do you know about innovaphone?", complete: true, src:"src0" });
messages.push({ role: "assistant", content: "innovaphone is a nice company!", complete: true, src:"src0" });
messages.push({ role: "user", content: "What exactly can you tell about ", src:"src0" });
messages.push({ role: "user", content: "their products?", complete: true, last:true, src:"src0" });
let seq = 0;
let reqBase = { api: "Assistant", mt: "Question" };
let src = new assistantApi.Src(result);
for (let i = 0; i < messages.length; i++) {
const req = { ...reqBase };
if (messages[i].content.length > WS_MAX_DATA_SIZE) {
for (let start = 0; start < messages[i].content.length; start += WS_MAX_DATA_SIZE) {
const chunk = messages[i].content.substring(start, start + WS_MAX_DATA_SIZE);
req.message = { role: messages[i].role, content: chunk };
req.seq = seq++;
if (start + WS_MAX_DATA_SIZE >= messages[i].content.length) {
req.complete = true;
if (i == messages.length - 1) req.last = true;
}
src.send(req);
}
}
else {
req.message = messages[i];
req.seq = seq++;
req.complete = true;
if (i == messages.length - 1) req.last = true;
src.send(req);
}
}
function result(recv) {
console.log(recv.msg?.message?.content);
src.close();
}