Smolagent-Chat CLI 😎

This is a simple chat interface to a Smolagents agentic search tool, inspired by the Smolagents multi-agents orchestration lesson at Huggingface.

It's an exercise in making a simple CLI tool that can be configured to use both local (Ollama) and remote (Anthropic) models and telemetry services (Arize/Phoenix). It's a great vehicle for learning.

Get Started

There are two ways to run this: directly in your local cli environment, or via docker

Local (no-docker)

Create a new conda forge environment:

conda create --prefix ./conda python=3.12
conda activate ./conda

Install dependencies

pip install -r requirements.txt

Docker

You can run the script in a docker container, using a read-only mount (for safety - it's good practice with agents)

docker build -t smolchat .
docker run --name smolchat --network local-net  -v ./:/app:ro -it smolchat python chat.py

then subsequent runs

docker start -i smolchat

NOTE If you are using ollama locally, add --network=host to the docker run command, also ensuring your ollama domain is host.docker.internal (for development only). If you're connecting to a local Phoenix container, change the network to --network local-net

Environment Configuration

Add your environment vars to .env (not tracked by git)

HF_TOKEN=your-huggingface-token

## ANTHROPIC ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
ANTHROPIC_TOKEN=your-anthropic-token
MODEL_ID="anthropic/claude-3-5-sonnet-20240620"

## OLLAMA / Locally ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# MODEL_ID="ollama_chat/qwen2.5-coder:7b-instruct"
# MODEL_ID="ollama_chat/granite3-dense:2b"
# LLM_API_BASE="http://host.docker.internal:11434"

## TELEMETRY ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

### Local / Docker
PHOENIX_COLLECTOR_ENDPOINT="http://phoenix:6006/v1/traces"

### Cloud
# PHOENIX_COLLECTOR_ENDPOINT=https://app.phoenix.arize.com
# PHOENIX_API_KEY=your-api-key
# PHOENIX_SPACE_ID=your-space-id

## WRAP UP ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
### Convenience
LLM_API_TOKEN=${ANTHROPIC_TOKEN}

Telemetry

This project assumes Arize / Phoenix telemetry is configured. Dependencies for both configurations are installed. Though smolagents also supports Langfuse, it is not included here.

You can run Phoenix locally via docker

First run

docker run --rm  --name phoenix --network local-net  -p 6006:6006 -p 4317:4317 -i -t arizephoenix/phoenix

subsequent runs

docker start -i phoenix

It's important to note that the docker-docker name of this container is phoenix, such that your collector endpoint should look like this in the .env file

PHOENIX_COLLECTOR_ENDPOINT=http://phoenix:6006/v1/traces

Or using a free Arize account, simply configure these in your .env

• PHOENIX_COLLECTOR_ENDPOINT
• PHOENIX_API_KEY
• PHOENIX_SPACE_ID

The script will configure local vs cloud based on the collector endpoint.

Local/Docker Environment Configuration

Docker:

PHOENIX_COLLECTOR_ENDPOINT="http://phoenix:6006/v1/traces"

Arize Cloud Environment Configuration

PHOENIX_COLLECTOR_ENDPOINT=https://app.phoenix.arize.com
PHOENIX_API_KEY=yourAPIkey
PHOENIX_SPACE_ID=FromArizeSpaceConfiguration

The Spaec ID can be obtained by creating a "New Project Or Model" from the project dashboard at arize.com