> For the complete documentation index, see [llms.txt](https://laixe.gitbook.io/laixe-documentation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://laixe.gitbook.io/laixe-documentation/basics/editor.md). # Welcome to Laixe Laixe is an AI project to replicate a KOL. Our ultimate goal is for the AI Agent to: * Identify promising tokens. * Announce them on Twitter and Telegram. * Auto-trade these tokens. Currently we only focus on Solana tokens, but in the future we will increase support for other chains. ### Telegram Channel Stat Tracking Methodology: 1. This is a static update which we will do every couple of days. 2. We scrape the messages from the last 90 days from the list of Telegram channels. 3. We record the timestamp and contract address/liquidity pool, since DEX Screener links have the liquidity pool in it. In case of liquidity pools, we replace it by identifying the contract address. 4. We then do a deduplication, where we consider the earliest instance of the contract address on the channel in the last 90 days. 5. We then record the highest price in the 5-minute candle of the token at the time of call, i.e. say if the token was called at 13:03:47 UTC on 1st January 2025, we consider the highest point in the candle from 13:00:00 and 13:04:59. This is to set a standard so that quick multipliers from copy traders are not taken into account. 6. We then record the highest price during/after this candle to record the high multiplier. 7. The 3x multiplier of a channel, which is a stat we track, is the number of calls that have at least a 3x multiplier divided by the total number of calls.

Fig 1: Flowchart for Backend Processes

### DEX Screener Boosts/Latest Tracking: We use the DEX Screener API to continuously record tokens which have boosted recently or have their token information updated, which normally happens on graduation to Raydium for pump fun tokens. ### DeepSeek Integration: We have integrated with DeepSeek V3 for the chatbot and DeepSeek R1 for the reasoning while scraping telegram channels live for notifications. We plan to integrate the data for latest tokens to the DeepSeek R1 model while recording information for it to identify ‘alpha’ tokens to call out on graduation to Raydium. Currently, DeepSeek R1 is also looking for mindshare amongst tokens in the telegram channels it tracks for live notifications, and it will look into calling out tokens with a high mindshare. Available functionality of text analysis on Telegram channels to identify tokens called out by KOLs with working technology and possibly viral or newsworthy events and tweeting them out on Twitter with the tag 'DeepSeek Recommendation'. ### Entertainment module: 1. Runs on a local machine of the developer with voice processing capabilities. 2. Uses Gemini speech-to-text API followed by DeepSeek chat for response, with a fallback to Gemini 1.5 pro model due to the unreliability of DeepSeek. 3. Runs the user response through Step 2 to get intent, and if the intent is detected as replying to a KOL within a list of \~100 KOLs, it replies to their latest tweet. In case of image generation, it generates an image using DeepSeek Janus Pro 7B and posts it on twitter. 4. Gives a voice response after processing the request. #### Further enhancements: * Custom voice using ElevenLabs API. * Video generation using LumaLabs. * AI streams on Youtube/Spaces/Twitch using AI avatars. * Rolling it out to users after understanding their requirements. ### Twitter Integration: We plan to integrate the DeepSeek R1 reasoning model for posting its calls on tokens to the official twitter automatically while integrating DeepSeek V3 for it to randomly post a few posts and reply to messages, while also interacting with other KOLs by raiding their posts. It will also shill its own token along with its crypto calls/bags. ### Strategy Identification: 1. Identify commonalities through machine learning on the successful telegram calls 2. Identifying tokens meeting those requirements for tokens graduating to raydium 3. Trying to identify mindshare of tokens on Telegram and Twitter while weighing their mindshare against the market cap ### AI Agent for Auto-trading: Once a strategy is successfully identified with successful calls, we will open up an auto-trading agent which users can use with a burn to use model with the $LAIXE token. Users will be able to set their filters, criteria for token selection with a text prompt, and take profits/stop losses after a test agent is deployed and tested using Laixe. ### Currently Available Features on Laixe: 1. **Query page:** A page where you can query for stats of telegram channels, linked to a backend SQL table 2. **DeepSeek Alerts:** Integrates DeepSeek V3 for the chatbot and also provides notifications on various telegram channels in the notifications feed. 3. **Social Scanner:** Use the Social Scanner page to look at the Twitter sentiment of tokens and gather information on them.