Copilot

Copilot helps users answer questions and assignments through different tools. So far, copilot supports two agents types: Langchain Agent and OpenAI Assistant Agent.

How to use Copilot as user

• Make sure docker is installed
• Get the etendo/etendo_copilot_core image from dockerhub: docker pull etendo/etendo_copilot_core:develop
• Once image is downloaded:
  • Set your local configuration copying .env.sample into .env and set the right values
  • Run a container as: docker run -it --env-file .env -p <host_machine_port>:<inside_docker_port> -v $(pwd)/:/app/ etendo/etendo_copilot_core:develop
  • Make a request sample: curl -i -X POST -H "Content-Type: application/json" -d '{"question": "What is etendo?"}' http://localhost:<host_machine_port>/question

Deploy docker image

This is done automatically from CI for develop and experimental branches.

docker build -t etendo/chatbot_etendo .
docker push etendo/chatbot_etendo

FRONT

kubectl port-forward -n chat-etendo svc/das 8092:8092
kubectl port-forward -n chat-etendo svc/etendo-retrieval 8085:8080

Backend Development

Virtual environment

As tool for managing multiple Python versions you can use pyenv. pyenv does not manage package dependencies, so for this purpose you can use Poetry. It will create an isolated virtual environment for the etendo-copilot-core project.

• Install pyenv: https://github.com/pyenv/pyenv#installation
• Install Poetry: curl -sSL https://install.python-poetry.org | python3 -
• Create etendo-copilot-core venv using poetry (recommended approach):

pyenv install 3.10
pyenv local 3.10
poetry env use 3.10
poetry install

Alternative, you can use Docker.

Torch and MacOS issue

If you are getting this issue from poetry install: Unable to find installation candidates for torch (2.0.1+cpu).

Workaround:

poetry shell
pip install torch==2.0.1
deactivate

Verify installation:

poetry run python
>>> import torch
>>> torch.__version__
'2.0.1'

Add new dependencies

• For prod dependency run: poetry add <dep_name>
• For dev dependency run: poetry add <dep_name> --group dev

Exporting dependencies

• To export dependencies to requirements.txt run: pip freeze > requirements.txt or poetry export -without-hashes -f requirements.txt > requirements.txt but removing the python versions range.

How to run copilot

• Locally outside docker:

  • Copy .env.sample into .env and set the right values
  • poetry run python run.py

• Using docker, make sure .env is created and all the variables are set, only then run docker run --env-file .env -p 5001:5001 etendo/chatbot_etendo. You can set the port that you want, just be sure to set the same port in the image from .env if not, the api will never be reached.

• The AGENT_TYPE environment variable should be used to set the agent type. There are two available agent: langchain and openai-assistant. By default copilot will be executed for langchain.

• Mount code as volume: docker run --env-file .env -p 5001:5001 -v $(pwd)/copilot:/app/copilot etendo/chatbot_etendo.

How to run unit test

poetry run pytest tests

OpenAPI URL

You can get the open api (swagger) documentation from http://localhost:<port>/docs or http://localhost:<port>/redoc

Pre-commit

• Install pre-commit from HERE
• Setup pre-commit pre-commit install & pre-commit autoupdate
• If you want to run for all the files: pre-commit run --all-files

Third Party Tools Implementation

Any developer can define his own tools and attach them into copilot agent. So as to do this the third party tools MUST be added into the tools package.

Baby steps to define a new tool from copilot source code

1- Create a new python module inside tools package: hello_world.py

2- Extend the ToolWrapper class from copilot.core.tool_wrapper and set your own tool implementation. Boilerplate sample:

from copilot.core.tool_wrapper import ToolWrapper

class MyTool(ToolWrapper):
    name = 'my_tool_name'
    description = 'My tool description'

    def __call__(self, *args, **kwargs):
        # Implement your tool's logic HERE

3- Enable the new tool from tools_config.json under third_party_tools:

{
    "native_tools": {
        ...
    },
    "third_party_tools": {
        "MyTool": true
    }
}

4- Restart the copilot container loading the project root folder through a volume: docker run --env-file .env -p 5001:5001 -v $(pwd):/app etendo/chatbot_etendo

Baby steps to define a new tool just using copilot image

1- Create a tools directory and inside it create a __init__.py file.

2- Create a new python module inside tools package: hello_world.py

3- Extend the ToolWrapper class from copilot.core.tool_wrapper and set your own tool implementation. Boilerplate sample:

from copilot.core.tool_wrapper import ToolWrapper

class MyTool(ToolWrapper):
    name = 'my_tool_name'
    description = 'My tool description'

    def __call__(self, *args, **kwargs):
        # Implement your tool's logic HERE

4- Expose the new tool class name from __init__.py

from .hello_world import MyTool

5- Enable the new tool from tools_config.json under third_party_tools:

{
    "native_tools": {
        ...
    },
    "third_party_tools": {
        "MyTool": true
    }
}

6- Restart the copilot container loading the project root folder through a volume: docker run --env-file .env -p 5001:5001 -v $(pwd)/tools:/app/tools -v $(pwd)/tools_config.json:/app/tools_config.json etendo/chatbot_etendo

Third Party Tools dependencies

Formats:

• pandas => Installing latest version
• pandas==1.3.3 => Installing a specific version
• pandas>=1.0.3 => Greater than or equal to a certain version
• pandas<=1.2.4 => Less than or equal to a certain version
• pandas>1.0.0 => Greater than a certain version
• pandas<2.0.0 => Less than a certain version
• pandas>=1.0.0,<=2.0.0 => Using version ranges
• pandas~=1.0.0 => Tilde operator (~) for installing compatible versions
• pandas^1.0.0 => Caret operator (^) for installing compatible versions

Resources

• Langchain Agents
• Assistants API
• How Assistants work