aos-rag

Intro

This project is designed to provide on-chain RAG(Retrieval-Augmented Generation) ability on top of aos process.

The target of the build is an ao module, everyone can start an aos process with this module loaded to achieve the ability to do on-chain RAG.

Quickstart

directly use our module(0FzCdvtr3yldxGsaTBkfdaSchhGwd-2k487ZoIogofI).

aos {process_name} --module 0FzCdvtr3yldxGsaTBkfdaSchhGwd-2k487ZoIogofI

Run .load command to load the test lua file.

1. Load Demo Code

.load process/rag.lua

2. Init Model & Vec DB

Send({ Target = ao.id, Action = "Init" })

Message received should be Database initialized

3. Embedding Data

Send({ Target = ao.id, Action = 'Embedding', Data = '["Shohei Ohtanis ex-interpreter pleads guilty to charges related to gambling and theft","The jury has been selected in Hunter Bidens gun trial","Larry Allen, a Super Bowl champion and famed Dallas Cowboy, has died at age 52","After saying Charlotte, a lone stingray, was pregnant, aquarium now says shes sick","An Epoch Times executive is facing money laundering charge"]' })

Message received should be Articles embedded

4. Retrieve

Send({ Target = ao.id, Action = "Retrieve", Data = '{"prompt":"firearm courtroom","limit":3}' })

Message received should be in JSON of sorted list contains headline & distance, distance represents how much headline is similarity with your prompt.

Contributing

Compile Module

// prepare vendor and make
// Makefile use docker image to build, so if there are difficulties with pulling image
// You can run docker pull to get the image by yourself

make vendor
make

Compilation Structure

Step 1:

• sqlite3.o : sqlite3.c
• libllama.a, common/libcommon.a : llamacpp cmake & make

Step 2:

• sqlite-vec.o: sqlite-vec.c -I{SQLITE_DIR} -DSQLITE_CORE
• sqlite-lembed.o: sqlite-lembed.c -I{SQLITE_DIR} -I{LLAMA_DIR} -DSQLITE_CORE

Step 3:

• lsqlite3.o: lsqlite3.c -I{SQLITE_VEC_DIR} -I{SQLITE_LEMBED_DIR} -I{LUA_LIB}

Step 4:

• libsqlite.so: sqlite3.o sqlite-vec.o sqlite-lembed.o lsqlite3.o

Step 5:

• Move libllama.a common/libcommon.a (from step 1) and libsqlite.so (from step 4) to aos/process/libs

Step 6:

• Ao build module to generate wasm

Publish Module

After running make, outputs process.js and process.wasm will be found in the directory test/.

// The size of output process.wasm may be 4.7M, it will cost 0.003 AR, there should be enough balanc in your AR wallet.

export WALLET='YOUR WALLET JSON FILE LOCATION'
node scripts/deploy_module.js

Reference

• AO-llama (https://github.com/PeterFarber/AO-Llama)
  This project demonstrates how to integrate the llama runtime environment into an AOS process as a callable library. It provides methods for compiling libraries within the process and outlines the basic compilation and linking logic for the llama library.

• sqlite-lembed (https://github.com/asg017/sqlite-lembed) This project provides the load extension function to support auto loading sqlite-lembed ability. After loaded, the library will provide functions in sqlite statements. We can use them in SQL statements.

• sqlite-vec (https://github.com/asg017/sqlite-vec) This project provides the load extension function to support auto loading sqlite-vec ability. After loaded, we can use the functions to process vetors and calculate the distance between vectors in SQL statements.