# PromethAI-Memory
AI Applications and RAGs - Cognitive Architecture, Testability, Production Ready Apps
Open-source framework for building and testing RAGs and Cognitive Architectures, designed for accuracy, transparency, and control.
[//]: # (Follow PromethAI
)
[//]: # ()
[//]: # ()
[//]: # (
)
[//]: # ()
[//]: # (
)
[//]: # (
)
[//]: # (
)
[//]: # (
)
[//]: # (
)
Share promethAI Repository
This repo is built to test and evolve RAG architecture, inspired by human cognitive processes, using Python. It's aims to be production ready, testable, but give great visibility in how we build RAG applications.
This project is a part of the [PromethAI](https://prometh.ai/) ecosystem.
It runs in iterations, with each iteration building on the previous one.
_Keep Ithaka always in your mind.
Arriving there is what you’re destined for.
But don’t hurry the journey at all.
Better if it lasts for years_
### Installation
To get started with PromethAI Memory, start with the latest iteration, and follow the instructions in the README.md file
### Current Focus
RAG test manager can be used via API or via the CLI
### Project Structure
#### Level 1 - OpenAI functions + Pydantic + DLTHub
Scope: Give PDFs to the model and get the output in a structured format
Blog post: https://prometh.ai/promethai-memory-blog-post-one
We introduce the following concepts:
- Structured output with Pydantic
- CMD script to process custom PDFs
#### Level 2 - Memory Manager + Metadata management
Scope: Give PDFs to the model and consolidate with the previous user activity and more
Blog post: https://prometh.ai/promethai-memory-blog-post-two
We introduce the following concepts:
- Long Term Memory -> store and format the data
- Episodic Buffer -> isolate the working memory
- Attention Modulators -> improve semantic search
- Docker
- API
#### Level 3 - Dynamic Graph Memory Manager + DB + Rag Test Manager
Scope: Store the data in N-related stores and test the retrieval with the Rag Test Manager
Blog post: https://prometh.ai/promethai-memory-blog-post-three
- Dynamic Memory Manager -> store the data in N hierarchical stores
- Auto-generation of tests
- Multiple file formats supported
- Postgres DB to store metadata
- Docker
- API
### Run the level 3
Make sure you have Docker, Poetry, and Python 3.11 installed and postgres installed.
Copy the .env.example to .env and fill in the variables
Two ways to run the level 3:
#### Docker:
Copy the .env.template to .env and fill in the variables
Specify the environment variable in the .env file to "docker"
Launch the docker image:
```docker compose up promethai_mem ```
Send the request to the API:
```
curl -X POST -H "Content-Type: application/json" -d '{
"payload": {
"user_id": "681",
"data": [".data/3ZCCCW.pdf"],
"test_set": "sample",
"params": ["chunk_size"],
"metadata": "sample",
"retriever_type": "single_document_context"
}
}' http://0.0.0.0:8000/rag-test/rag_test_run
```
Params:
data -> list of URLs or path to the file, located in the .data folder (pdf, docx, txt, html)
test_set -> sample, manual (list of questions and answers)
metadata -> sample, manual (json) or version (in progress)
params -> chunk_size, chunk_overlap, search_type (hybrid, bm25), embeddings
retriever_type -> llm_context, single_document_context, multi_document_context, cognitive_architecture(coming soon)
Inspect the results in the DB:
``` docker exec -it postgres psql -U bla ```
``` \c bubu ```
``` select * from test_outputs; ```
Or set up the superset to visualize the results:
#### Poetry environment:
Copy the .env.template to .env and fill in the variables
Specify the environment variable in the .env file to "local"
Use the poetry environment:
``` poetry shell ```
Change the .env file Environment variable to "local"
Launch the postgres DB
``` docker compose up postgres ```
Launch the superset
``` docker compose up superset ```
Open the superset in your browser
``` http://localhost:8088 ```
Add the Postgres datasource to the Superset with the following connection string:
``` postgres://bla:bla@postgres:5432/bubu ```
Make sure to run to initialize DB tables
``` python scripts/create_database.py ```
After that, you can run the RAG test manager from your command line.
```
python rag_test_manager.py \
--file ".data" \
--test_set "example_data/test_set.json" \
--user_id "666" \
--metadata "example_data/metadata.json" \
--retriever_type "single_document_context"
```
Examples of metadata structure and test set are in the folder "example_data"