MAESTRO Core

This is Minimal subset of MAESTRO to demonstrate MAESTRO architecture.

Framework description is available here: https://airi-institute.github.io/maestro-cover

It consist of:

• gateway :: backend gateway for all requests
• chat-manager-examples :: component which manages bots business-logic
• llm-hub :: service to interact with LLM

Usage: steps

Start MAESTRO:

make build setup-env up

Check basic track:

make run-dummy records='dummy=hello dummy=test dummy=exit'

Setup LLM

LLM integrations supported. (Right now only OpenRouter or GigaChat)

If you have keys, you need to setup llm_config.json, configuration for LLM.

Variant 1

• copy llm_config.json.example to llm_config.json
• edit llm_config.json :: fill ??? placeholders with your keys, fix model_id if you need, remove excess llm_config
• copy llm_config.json to ./data, run make update-llm-config
• restart llm-hub, run make restart-llm-hub

Variant 2

• use wizard to setup llm_config.json: run make run-wizard
• follow MAESTRO CLI steps.
• copy llm_config.json to ./data, run make update-llm-config
• restart llm-hub, run make restart-llm-hub

Run LLM example

make run-chatbot records='start="Какая ты языковая модель?"'

Questions forwarded to your configured LLM.

Run Describer example

make run-describer

document-extractor (EXPERIMENTAL)

Build and up container with document-extractor

• on CPU :: make document-extractor-up
• on GPU :: make document-extractor-up-on-gpu
  • assumed CUDA with version >= 12.4 available
  • run nvidia-smi | grep -o 'CUDA Version.*' to check it

Prepare your document

make prepare-document

Run DocumentDescriber example

make run-document-describer

Note: document-extractor works pretty slow on CPU.

frontend-telegram

Create chatbot

Via @BotFather: https://t.me/BotFather

Setup .env

Update ./data/.env:

TG_APPLICATION__HANDLE=@your-bot-handle
TG_APPLICATION__TOKEN=your-bot-token
bot__commands={"start": "Dummy"}

AUTH__TG_PASSWORD=password-to-yourbot
# optional

Run

docker compose --file=compose--frontend-telegram.yaml --file=compose.yaml up