agently
Version:
🤵 Agently is a framework helps developers to create amazing LLM based applications. 🎭 You can use it to create an LLM bansed agent instance with role set and memory easily. ⚙️ You can use Agently agent instance just like an async function and put it any
316 lines (312 loc) • 12 kB
JavaScript
const EventEmitter = require('events'),
axios = require('axios-https-proxy-fix')
module.exports = (Agently) => {
//OpenAI GPT 3.5 turbo 4k
Agently.LLM.Manage
//Model Name
.name('GPT')
//API URL
.url('https://api.openai.com/v1/chat/completions')
//Default Options
.defaultOptions({
model: 'gpt-3.5-turbo',
temperature: 0.8,
})
//Default Max Context Length (Limited by model runtime tokens limitation)
.defaultMaxContextLength(2500)
//Request Method
.request(
async (reqData) => {
let data = { ...reqData.options } || {}
data.model = data.model || 'gpt-3.5-turbo'
data.messages = reqData.messages
data.stream = false
const result = await axios.request({
url: reqData.url,
method: 'POST',
data: data,
headers: {
'Authorization': `Bearer ${ reqData.auth }`,
'Content-Type': 'application/json',
},
dataType: 'json',
proxy:
reqData.proxy?.host && reqData.proxy?.port
? {
host: reqData.proxy.host,
port: reqData.proxy.port,
}
: undefined
})
return result.data
}
)
//Extract response data
.extractResponse(
res => res.data.choices[0].message.content
)
//Request in streaming way
.streaming(
async (reqData) => {
let data = { ...reqData.options } || {}
data.model = data.model || 'gpt-3.5-turbo'
data.messages = reqData.messages
data.stream = true
const result = await axios.request({
url: reqData.url,
method: 'POST',
data: data,
headers: {
'Authorization': `Bearer ${ reqData.auth }`,
'Content-Type': 'application/json',
},
dataType: 'json',
responseType: 'stream',
proxy:
reqData.proxy?.host && reqData.proxy?.port
? {
host: reqData.proxy.host,
port: reqData.proxy.port,
}
: undefined
})
return result.data
}
)
//Extract streaming data
//when streaming is still on: return { type: 'data', data: <choices 0 content> }
//when streaming is done: return { type: 'done', data: null }
.extractStreamingData(
data => {
if (data && data !== '[DONE]\n\n') {
try {
return { type: 'data', data: JSON.parse(data)?.choices[0] || '' }
} catch (e) {
return { type: 'data', data: '' }
//console.error(e)
}
} else {
return { type: 'done', data: null }
}
}
)
.register()
//OpenAI GPT 3.5 turbo 16k
Agently.LLM.Manage
//Model Name
.name('GPT-16K')
//API URL
.url('https://api.openai.com/v1/chat/completions')
//Default Options
.defaultOptions({
model: 'gpt-3.5-turbo-16k',
temperature: 0.8,
})
//Default Max Context Length (Limited by model runtime tokens limitation)
.defaultMaxContextLength(13500)
//Request Method
.request(
async (reqData) => {
let data = { ...reqData.options } || {}
data.model = data.model || 'gpt-3.5-turbo-16k'
data.messages = reqData.messages
data.stream = false
const result = await axios.request({
url: reqData.url,
method: 'POST',
data: data,
headers: {
'Authorization': `Bearer ${ reqData.auth }`,
'Content-Type': 'application/json',
},
dataType: 'json',
proxy:
reqData.proxy?.host && reqData.proxy?.port
? {
host: reqData.proxy.host,
port: reqData.proxy.port,
}
: undefined
})
return result.data
}
)
//Extract response data
.extractResponse(
res => res.data.choices[0].message.content
)
//Request in streaming way
.streaming(
async (reqData) => {
let data = { ...reqData.options } || {}
data.model = data.model || 'gpt-3.5-turbo-16k'
data.messages = reqData.messages
data.stream = true
const result = await axios.request({
url: reqData.url,
method: 'POST',
data: data,
headers: {
'Authorization': `Bearer ${ reqData.auth }`,
'Content-Type': 'application/json',
},
dataType: 'json',
responseType: 'stream',
proxy:
reqData.proxy?.host && reqData.proxy?.port
? {
host: reqData.proxy.host,
port: reqData.proxy.port,
}
: undefined
})
return result.data
}
)
//Extract streaming data
//when streaming is still on: return { type: 'data', data: <choices 0 content> }
//when streaming is done: return { type: 'done', data: null }
.extractStreamingData(
data => {
if (data && data !== '[DONE]\n\n') {
try {
return { type: 'data', data: JSON.parse(data)?.choices[0] || '' }
} catch (e) {
return { type: 'data', data: '' }
//console.error(e)
}
} else {
return { type: 'done', data: null }
}
}
)
.register()
//MiniMax (default model: abab5.5-chat)
Agently.LLM.Manage
//Model Name
.name('MiniMax')
//API URL
.url('https://api.minimax.chat/v1/text/chatcompletion')
//Default Options
.defaultOptions({
model: 'abab5-chat',
tokens_to_generate: 3000,
temperature: 0.8,
})
//Request Method
.request(
async (reqData) => {
let data = { ...reqData.options } || {}
data.model = data.model || 'abab5-chat'
data.stream = false
//Split SYSTEM messages to prompt
let prompt = ''
data.messages = []
for (let i = 0; i < reqData.messages.length; i++) {
if (reqData.messages[i]['sender_type'] === 'SYSTEM') {
prompt += `${ reqData.messages[i].text }\n`
} else {
data.messages.push(reqData.messages[i])
}
}
if (prompt !== '') {
data.prompt = prompt
data.role_meta = {
user_name: reqData.userName || 'USER',
bot_name: reqData.botName || 'BOT',
}
}
const result = await axios.request({
url: `${ reqData.url }?GroupId=${ reqData.auth.groupId }`,
method: 'POST',
data: data,
headers: {
'Authorization': `Bearer ${ reqData.auth.apiKey }`,
'Content-Type': 'application/json',
},
dataType: 'json',
})
return result.data
}
)
//Extract response data
.extractResponse(
res => (
res.data.choices
? res.data.choices[0].text
: `Error: ${ JSON.stringify(res.data.base_resp.status_msg) }`
)
)
//Request in streaming way
.streaming(
async (reqData) => {
let data = { ...reqData.options } || {}
data.model = 'abab5.5-chat'
data.stream = true
data.use_standard_sse = false
//Split SYSTEM messages from messages and put it into prompt
let prompt = ''
data.messages = []
for (let i = 0; i < reqData.messages.length; i++) {
if (reqData.messages[i]['sender_type'] === 'SYSTEM') {
prompt += `${ reqData.messages[i].text }\n`
} else {
data.messages.push(reqData.messages[i])
}
}
if (prompt !== '') {
data.prompt = prompt
data.role_meta = {
user_name: reqData.userName || 'USER',
bot_name: reqData.botName || 'BOT',
}
}
const result = await axios.request({
url: `${ reqData.url }?GroupId=${ reqData.auth.groupId }`,
method: 'POST',
data: data,
headers: {
'Authorization': `Bearer ${ reqData.auth.apiKey }`,
'Content-Type': 'application/json',
},
dataType: 'json',
responseType: 'stream',
})
return result.data
}
)
//Extract streaming data
//when streaming is still on: return { type: 'data', data: <choices 0 content> }
//when streaming is done: return { type: 'done', data: null }
.extractStreamingData(
data => {
data = JSON.parse(data)
if (data.reply && data.choices[0].delta === '') {
return { type: 'done', data: null }
} else {
data.choices[0].delta = { content: data.choices[0].delta }
return { type: 'data', data: data.choices[0] }
}
}
)
//If model required messages queue format is not the same as GPT messages queue,
//we need to define how to transform the messages into the format model need.
.transformMessages(
messages => {
if (!(messages instanceof Array)) messages = [messages]
const roleDict = {
'system': 'SYSTEM',
'user': 'USER',
'assistant': 'BOT',
}
return messages.map(
item => {
return { sender_type: roleDict[item.role.toLowerCase()], text: item.content }
}
)
}
)
.register()
//Set default LLM in a new agent
Agently.Options.set('defaultAgentSettings.llmName', 'GPT')
}